drm/amdgpu/pm: support mca_ceumc_addr in ecctable
Subsystems (3)
PATCH 1/2 drm/amdgpu/pm: support mca_ceumc_addr in ecctable
SMU add a new variable mca_ceumc_addr to record umc correctable error address in EccInfo table, driver side add ecctable_v2 to support this feature
--- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h | 1 + drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h | 2 + .../inc/pmfw_if/smu13_driver_if_aldebaran.h | 15 +++ .../drm/amd/pm/swsmu/smu13/aldebaran_ppt.c | 101 ++++++++++++++---- .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c | 2 + 5 files changed, 98 insertions(+), 23 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h +1 -0
@@ -328,6 +328,7 @@ struct ecc_info_per_ch {uint16_t ce_count_hi_chip;uint64_t mca_umc_status;uint64_t mca_umc_addr; +uint64_t mca_ceumc_addr; };struct umc_ecc_info {
drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h +2 -0
@@ -322,6 +322,7 @@ enum smu_table_idSMU_TABLE_PACE ,SMU_TABLE_ECCINFO ,SMU_TABLE_COMBO_PPTABLE , +SMU_TABLE_ECCINFO_V2 , SMU_TABLE_COUNT, };
@@ -340,6 +341,7 @@ struct smu_table_contextvoid *driver_pptable;void *combo_pptable;void *ecc_table; +void *ecc_table_v2;// adapt to smu support record mca_ceumc_addr void *driver_smu_config_table;struct smu_table tables[SMU_TABLE_COUNT ];/*
drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu13_driver_if_aldebaran.h +15 -0
@@ -522,6 +522,21 @@ typedef struct {EccInfo_t EccInfo[ALDEBARAN_UMC_CHANNEL_NUM ]; } EccInfoTable_t; +typedef struct { +uint64_t mca_umc_status; +uint64_t mca_umc_addr; +uint64_t mca_ceumc_addr; + +uint16_t ce_count_lo_chip; +uint16_t ce_count_hi_chip; + +uint32_t eccPadding; +}EccInfo_t_v2 ; + +typedef struct { +EccInfo_t_v2 EccInfo[ALDEBARAN_UMC_CHANNEL_NUM ]; +}EccInfoTable_t_v2 ; +// These defines are used with the following messages: // SMC_MSG_TransferTableDram2Smu // SMC_MSG_TransferTableSmu2Dram
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +78 -23
145 lines not on this page · raw message
drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c +2 -0
@@ -552,9 +552,11 @@ int smu_v13_0_fini_smc_tables(struct smu_context *smu)kfree (smu_table->hardcode_pptable); smu_table->hardcode_pptable =NULL ; +kfree (smu_table->ecc_table_v2);kfree (smu_table->ecc_table);kfree (smu_table->metrics_table);kfree (smu_table->watermarks_table); + smu_table->ecc_table_v2 =NULL ; smu_table->ecc_table =NULL ; smu_table->metrics_table =NULL ; smu_table->watermarks_table =NULL ;
Signature · 1 line
2.17.1
PATCH 2/2 drm/amdgpu: print umc correctable error address
--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 5 ++ drivers/gpu/drm/amd/amdgpu/umc_v6_7.c | 55 ++++++++++++++++++- .../drm/amd/pm/swsmu/smu13/aldebaran_ppt.c | 2 + 3 files changed, 60 insertions(+), 2 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu.h +5 -0
@@ -1108,6 +1108,11 @@ struct amdgpu_device {bool scpm_enabled;uint32_t scpm_status; + +/* Determine smu ecctable whether support +* record correctable error address +*/ +int record_ce_addr_supported; };static inline struct amdgpu_device *drm_to_adev (struct drm_device *ddev)
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +53 -2
79 lines not on this page · raw message
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +2 -0
@@ -1908,6 +1908,8 @@ static ssize_t aldebaran_get_ecc_info(struct smu_context *smu, ecc_table_v2->EccInfo[i].mca_umc_addr; ecc_info_per_channel->mca_ceumc_addr = ecc_table_v2->EccInfo[i].mca_ceumc_addr; +if (!smu->adev->record_ce_addr_supported) + smu->adev->record_ce_addr_supported =1 ; } }
Signature · 1 line
2.17.1
RE: [PATCH Review 2/2] drm/amdgpu: print umc correctable error address
[AMD Official Use Only - General]
Quoted 67 lines
-----Original Message-----
From: Stanley.Yang <[email protected]>
Sent: Monday, May 23, 2022 4:17 PM
To: [email protected]; Zhang, Hawking <[email protected]>; Zhou1, Tao <[email protected]>; Quan, Evan <[email protected]>; Lazar, Lijo <[email protected]>
Cc: Yang, Stanley <[email protected]>
Subject: [PATCH Review 2/2] drm/amdgpu: print umc correctable error addressSigned-off-by Stanley.Yang <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 5 ++ drivers/gpu/drm/amd/amdgpu/umc_v6_7.c | 55 ++++++++++++++++++- .../drm/amd/pm/swsmu/smu13/aldebaran_ppt.c | 2 + 3 files changed, 60 insertions(+), 2 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu.h +5
-0@@ -1108,6 +1108,11 @@ struct amdgpu_device {bool scpm_enabled;uint32_t scpm_status; + +/* Determine smu ecctable whether support +* record correctable error address +*/ +int record_ce_addr_supported; };static inline struct amdgpu_device *drm_to_adev (struct drm_device *ddev)drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +0
-0@@ -119,6 +119,27 @@ static voidumc_v6_7_ecc_info_query_correctable_error_count(struct
amdgpu_devicedrivers/gpu/drm/amd/amdgpu/umc_v6_7.c +0
-0*error_count +=1 ; umc_v6_7_query_error_status_helper(adev,
mc_umc_status,umc_reg_offset);drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +5
-0+ +if (adev->record_ce_addr_supported) { +uint64_t err_addr, soc_pa; +uint32_t channel_index = + adev->umc.channel_idx_tbl[umc_inst *
adev->umc.channel_inst_num+drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +3
-0+ch_inst]; + + err_addr = ras-Quoted 1 line
umc_ecc.ecc[eccinfo_table_idx].mca_ceumc_addr;
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +1
-0+ err_addr = REG_GET_FIELD(err_addr,
MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +1
-0+/* translate umc channel address to soc pa, 3 parts are included */
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +2
-0+ soc_pa = ADDR_OF_8KB_BLOCK(err_addr) | +ADDR_OF_256B_BLOCK(channel_index) |drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +3
-0+OFFSET_IN_256B_BLOCK (err_addr); + +/* The umc channel bits are not original values, they are hashed */
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c static void umc_v6_7_ecc_info_query_correctable_error_count(struct amdgpu_device · line 139
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +4
-0+SET_CHANNEL_HASH (channel_index, soc_pa); + +/* clear [C4 C3 C2] in soc physical address */ + soc_pa &= ~(0x7ULL <<UMC_V6_7_PA_C2_BIT );
[Tao] this clear is the preparation for looping all column bits in same row, you only need physical address of one page, the code can be removed.
Quoted 86 lines
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +2
-0+ + dev_info(adev->dev, "Error Address(PA): 0x%llx\n",
soc_pa);drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +1
-0+ } } }@@ -251,7 +272,9 @@ static voidumc_v6_7_ecc_info_query_ras_error_address(struct
amdgpu_device*adevstatic void umc_v6_7_query_correctable_error_count(struct
amdgpu_device*adev,drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +0
-1uint32_t umc_reg_offset, - unsigned long
*error_count)drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +1
-0+unsigned long
*error_count,drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +2
-0+uint32_t ch_inst, + uint32_t umc_inst) { uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr; uint32_t ecc_err_cnt, ecc_err_cnt_addr; @@ -295,6 +318,33 @@static void umc_v6_7_query_correctable_error_count(struct
amdgpu_device*adev,drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +0
-0*error_count +=1 ; umc_v6_7_query_error_status_helper(adev,
mc_umc_status,umc_reg_offset);drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +8
-0+ + { +uint64_t err_addr, soc_pa; +uint32_t mc_umc_addrt0; +uint32_t channel_index; + + mc_umc_addrt0 = + SOC15_REG_OFFSET(UMC, 0,
regMCA_UMC_UMC0_MCUMC_ADDRT0);drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +3
-0+ + channel_index = + adev->umc.channel_idx_tbl[umc_inst *
adev->umc.channel_inst_num+drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +3
-0+ch_inst]; + + err_addr = RREG64_PCIE((mc_umc_addrt0 +
umc_reg_offset) * 4);drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +1
-0+ err_addr = REG_GET_FIELD(err_addr,
MCA_UMC_UMC0_MCUMC_ADDRT0,drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +3
-0+ErrorAddr); + +/* translate umc channel address to soc pa, 3 parts are included */
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +2
-0+ soc_pa = ADDR_OF_8KB_BLOCK(err_addr) | +ADDR_OF_256B_BLOCK(channel_index) |drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +3
-0+OFFSET_IN_256B_BLOCK (err_addr); + +/* The umc channel bits are not original values, they are hashed */
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +5
-0+SET_CHANNEL_HASH (channel_index, soc_pa); + +/* clear [C4 C3 C2] in soc physical address */ + soc_pa &= ~(0x7ULL <<UMC_V6_7_PA_C2_BIT ); + dev_info(adev->dev, "Error Address(PA): 0x%llx\n",
soc_pa);drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +3
-1+ } } }@@ -395,7 +445,8 @@ static void umc_v6_7_query_ras_error_count(structamdgpu_device *adev, ch_inst);umc_v6_7_query_correctable_error_count (adev, umc_reg_offset, - &(err_data->ce_count)); + &(err_data->ce_count), + ch_inst, umc_inst); umc_v6_7_querry_uncorrectable_error_count(adev, umc_reg_offset,&(err_data-
Quoted 1 line
ue_count));drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +0
-0
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c static ssize_t aldebaran_get_ecc_info(struct · line 1912
Quoted 5 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +1
-0@@ -1908,6 +1908,8 @@ static ssize_t aldebaran_get_ecc_info(struct smu_context *smu, ecc_table_v2->EccInfo[i].mca_umc_addr; ecc_info_per_channel->mca_ceumc_addr = ecc_table_v2->EccInfo[i].mca_ceumc_addr; +if (!smu->adev->record_ce_addr_supported)
[Tao] it seems the check is unnecessary.
Quoted 6 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +1
-1+ smu->adev->record_ce_addr_supported =1 ; } } --2.17.1
回复: [PATCH Review 2/2] drm/amdgpu: print umc correctable error address
Thanks tao, will update.
Quoted 189 lines
-----邮件原件-----
发件人: Zhou1, Tao <[email protected]>
发送时间: Monday, May 23, 2022 6:22 PM
收件人: Yang, Stanley <[email protected]>; amd- [email protected]; Zhang, Hawking <[email protected]>; Quan, Evan <[email protected]>; Lazar, Lijo <[email protected]>
抄送: Yang, Stanley <[email protected]>
主题: RE: [PATCH Review 2/2] drm/amdgpu: print umc correctable error address[AMD Official Use Only - General]
Quoted 5 lines
-----Original Message-----
From: Stanley.Yang <[email protected]>
Sent: Monday, May 23, 2022 4:17 PM
To: [email protected]; Zhang, Hawking <[email protected]>; Zhou1, Tao <[email protected]>; Quan,Evan
Quoted 62 lines
<[email protected]>; Lazar, Lijo <[email protected]>
Cc: Yang, Stanley <[email protected]>
Subject: [PATCH Review 2/2] drm/amdgpu: print umc correctable error addressSigned-off-by Stanley.Yang <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu.h | 5 ++ drivers/gpu/drm/amd/amdgpu/umc_v6_7.c | 55 ++++++++++++++++++- .../drm/amd/pm/swsmu/smu13/aldebaran_ppt.c | 2 + 3 files changed, 60 insertions(+), 2 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu.h +5
-0@@ -1108,6 +1108,11 @@ struct amdgpu_device {bool scpm_enabled;uint32_t scpm_status; + +/* Determine smu ecctable whether support +* record correctable error address +*/ +int record_ce_addr_supported; }; static inline struct amdgpu_device *drm_to_adev(struct drm_device*ddev) diff --git
a/drivers/gpu/drm/amd/amdgpu/umc_v6_7.cb/drivers/gpu/drm/amd/amdgpu/umc_v6_7.c
index606892dbea1c..47bd39e52e9b100644drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +0
-0@@ -119,6 +119,27 @@ static voidumc_v6_7_ecc_info_query_correctable_error_count(struct
amdgpu_devicedrivers/gpu/drm/amd/amdgpu/umc_v6_7.c +0
-0*error_count +=1 ; umc_v6_7_query_error_status_helper(adev,
mc_umc_status,umc_reg_offset);drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +5
-0+ +if (adev->record_ce_addr_supported) { +uint64_t err_addr, soc_pa; +uint32_t channel_index = + adev->umc.channel_idx_tbl[umc_inst *
adev->umc.channel_inst_num+drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +3
-0+ch_inst]; + + err_addr = ras-Quoted 1 line
umc_ecc.ecc[eccinfo_table_idx].mca_ceumc_addr;
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +1
-0+ err_addr = REG_GET_FIELD(err_addr,
MCA_UMC_UMC0_MCUMC_ADDRT0, ErrorAddr);drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +1
-0+/* translate umc channel address to soc pa, 3 parts are included */
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +2
-0+ soc_pa = ADDR_OF_8KB_BLOCK(err_addr) | +ADDR_OF_256B_BLOCK(channel_index) |drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +3
-0+OFFSET_IN_256B_BLOCK (err_addr); + +/* The umc channel bits are not original values, they are hashed */
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c static void umc_v6_7_ecc_info_query_correctable_error_count(struct amdgpu_device · line 139
Quoted 4 lines
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +4
-0+SET_CHANNEL_HASH (channel_index, soc_pa); + +/* clear [C4 C3 C2] in soc physical address */ + soc_pa &= ~(0x7ULL <<UMC_V6_7_PA_C2_BIT );[Tao] this clear is the preparation for looping all column bits in same row, you only need physical address of one page, the code can be removed.
Quoted 86 lines
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +2
-0+ + dev_info(adev->dev, "Error Address(PA): 0x%llx\n",
soc_pa);drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +1
-0+ } } }@@ -251,7 +272,9 @@ static voidumc_v6_7_ecc_info_query_ras_error_address(struct
amdgpu_device*adevstatic void umc_v6_7_query_correctable_error_count(struct
amdgpu_device*adev,drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +0
-1uint32_t umc_reg_offset, - unsigned long
*error_count)drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +1
-0+unsigned long
*error_count,drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +2
-0+uint32_t ch_inst, + uint32_t umc_inst) { uint32_t ecc_err_cnt_sel, ecc_err_cnt_sel_addr; uint32_t ecc_err_cnt, ecc_err_cnt_addr; @@ -295,6 +318,33 @@ staticvoid umc_v6_7_query_correctable_error_count(struct
amdgpu_device*adev,drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +0
-0*error_count +=1 ; umc_v6_7_query_error_status_helper(adev,
mc_umc_status,umc_reg_offset);drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +8
-0+ + { +uint64_t err_addr, soc_pa; +uint32_t mc_umc_addrt0; +uint32_t channel_index; + + mc_umc_addrt0 = + SOC15_REG_OFFSET(UMC, 0,
regMCA_UMC_UMC0_MCUMC_ADDRT0);drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +3
-0+ + channel_index = + adev->umc.channel_idx_tbl[umc_inst *
adev->umc.channel_inst_num+drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +3
-0+ch_inst]; + + err_addr = RREG64_PCIE((mc_umc_addrt0 +
umc_reg_offset) * 4);drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +1
-0+ err_addr = REG_GET_FIELD(err_addr,
MCA_UMC_UMC0_MCUMC_ADDRT0,drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +3
-0+ErrorAddr); + +/* translate umc channel address to soc pa, 3 parts are included */
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +2
-0+ soc_pa = ADDR_OF_8KB_BLOCK(err_addr) | +ADDR_OF_256B_BLOCK(channel_index) |drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +3
-0+OFFSET_IN_256B_BLOCK (err_addr); + +/* The umc channel bits are not original values, they are hashed */
drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +5
-0+SET_CHANNEL_HASH (channel_index, soc_pa); + +/* clear [C4 C3 C2] in soc physical address */ + soc_pa &= ~(0x7ULL <<UMC_V6_7_PA_C2_BIT ); + dev_info(adev->dev, "Error Address(PA): 0x%llx\n",
soc_pa);drivers/gpu/drm/amd/amdgpu/umc_v6_7.c +3
-1+ } } }@@ -395,7 +445,8 @@ static void umc_v6_7_query_ras_error_count(structamdgpu_device *adev, ch_inst);umc_v6_7_query_correctable_error_count (adev, umc_reg_offset, - &(err_data->ce_count)); + &(err_data->ce_count), + ch_inst, umc_inst); umc_v6_7_querry_uncorrectable_error_count(adev, umc_reg_offset,&(err_data-
Quoted 1 line
ue_count));drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +0
-0drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c static ssize_t aldebaran_get_ecc_info(struct · line 1912
Quoted 5 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +1
-0@@ -1908,6 +1908,8 @@ static ssize_t aldebaran_get_ecc_info(struct smu_context *smu, ecc_table_v2->EccInfo[i].mca_umc_addr; ecc_info_per_channel->mca_ceumc_addr = ecc_table_v2->EccInfo[i].mca_ceumc_addr; +if (!smu->adev->record_ce_addr_supported)[Tao] it seems the check is unnecessary.
Quoted 6 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +1
-1+ smu->adev->record_ce_addr_supported =1 ; } } --2.17.1
[AMD Official Use Only - General]
________________________________
From: amd-gfx <[email protected]> on behalf of Stanley.Yang <[email protected]>
Sent: Monday, May 23, 2022 4:17 PM
To: [email protected] <[email protected]>; Zhang, Hawking <[email protected]>; Zhou1, Tao <[email protected]>; Quan, Evan <[email protected]>; Lazar, Lijo <[email protected]>
Cc: Yang, Stanley <[email protected]>
Subject: [PATCH Review 1/2] drm/amdgpu/pm: support mca_ceumc_addr in ecctable
SMU add a new variable mca_ceumc_addr to record umc correctable error address in EccInfo table, driver side add ecctable_v2 to support this feature
--- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h | 1 + drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h | 2 + .../inc/pmfw_if/smu13_driver_if_aldebaran.h | 15 +++ .../drm/amd/pm/swsmu/smu13/aldebaran_ppt.c | 101 ++++++++++++++---- .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c | 2 + 5 files changed, 98 insertions(+), 23 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h +1 -0
@@ -328,6 +328,7 @@ struct ecc_info_per_ch {uint16_t ce_count_hi_chip;uint64_t mca_umc_status;uint64_t mca_umc_addr; +uint64_t mca_ceumc_addr; };struct umc_ecc_info {
drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h +2 -0
@@ -322,6 +322,7 @@ enum smu_table_idSMU_TABLE_PACE ,SMU_TABLE_ECCINFO ,SMU_TABLE_COMBO_PPTABLE , +SMU_TABLE_ECCINFO_V2 , SMU_TABLE_COUNT, };
@@ -340,6 +341,7 @@ struct smu_table_contextvoid *driver_pptable;void *combo_pptable;void *ecc_table; +void *ecc_table_v2;// adapt to smu support record mca_ceumc_addr void *driver_smu_config_table;struct smu_table tables[SMU_TABLE_COUNT ];/*
drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu13_driver_if_aldebaran.h +15 -0
@@ -522,6 +522,21 @@ typedef struct {EccInfo_t EccInfo[ALDEBARAN_UMC_CHANNEL_NUM ]; } EccInfoTable_t; +typedef struct { +uint64_t mca_umc_status; +uint64_t mca_umc_addr; +uint64_t mca_ceumc_addr; + +uint16_t ce_count_lo_chip; +uint16_t ce_count_hi_chip; + +uint32_t eccPadding; +}EccInfo_t_v2 ; + +typedef struct { +EccInfo_t_v2 EccInfo[ALDEBARAN_UMC_CHANNEL_NUM ]; +}EccInfoTable_t_v2 ; +// These defines are used with the following messages: // SMC_MSG_TransferTableDram2Smu // SMC_MSG_TransferTableSmu2Dram
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +9 -0
@@ -82,6 +82,12 @@*/ #define SUPPORT_ECCTABLE_SMU_VERSION 0x00442a00 +/* +* SMU support mca_ceumc_addr in ECCTABLE since version 68.55.0, +* use this to check mca_ceumc_addr record whether support +*/ +#define SUPPORT_ECCTABLE_V2_SMU_VERSION 0x00443700 +/* * SMU support BAD CHENNEL info MSG since version 68.51.00, * use this to check ECCTALE feature whether support
@@ -239,6 +245,9 @@ static int aldebaran_tables_init(struct smu_context *smu)SMU_TABLE_INIT (tables,SMU_TABLE_ECCINFO ,sizeof (EccInfoTable_t),PAGE_SIZE ,AMDGPU_GEM_DOMAIN_VRAM ); +SMU_TABLE_INIT (tables,SMU_TABLE_ECCINFO_V2 ,sizeof (EccInfoTable_t_v2), +PAGE_SIZE ,AMDGPU_GEM_DOMAIN_VRAM ); +
[kevin]:
this table mapping is not needed, the reason as below.
smu_table->metrics_table = kzalloc(sizeof(SmuMetrics_t), GFP_KERNEL);
if (!smu_table->metrics_table)
return -ENOMEM;
@@ -255,6 +264,10 @@ static int aldebaran_tables_init(struct smu_context *smu)
if (!smu_table->ecc_table)
return -ENOMEM;+ smu_table->ecc_table_v2 = kzalloc(tables[SMU_TABLE_ECCINFO_V2].size, GFP_KERNEL);
+ if (!smu_table->ecc_table_v2)
+ return -ENOMEM;;
+
[kevin]:
add eccinfo table v2 is not needed for this case, this table is only used store table data from pmfw to driver, you can create a large enough table which can save ecc table data directly.
e.g:
size = max(sizeof(EccInfoTable_t_v2), sizeof(EccInfoTable_t)); smu_table->ecc_table = kzalloc(size, GFP_KERNEL);
Best Regards,
Kevin
return 0;
}
@@ -1802,7 +1815,8 @@ static ssize_t aldebaran_get_gpu_metrics(struct smu_context *smu,
return sizeof(struct gpu_metrics_v1_3);
}
-static int aldebaran_check_ecc_table_support(struct smu_context *smu) +static int aldebaran_check_ecc_table_support(struct smu_context *smu,
+ int *ecctable_version)
{
uint32_t if_version = 0xff, smu_version = 0xff;
int ret = 0;@@ -1815,6 +1829,11 @@ static int aldebaran_check_ecc_table_support(struct smu_context *smu)
if (smu_version < SUPPORT_ECCTABLE_SMU_VERSION)
ret = -EOPNOTSUPP;
+ else if (smu_version >= SUPPORT_ECCTABLE_SMU_VERSION &&
+ smu_version < SUPPORT_ECCTABLE_V2_SMU_VERSION)
+ *ecctable_version = 1;
+ else
+ *ecctable_version = 2;
return ret;
}
@@ -1824,36 +1843,72 @@ static ssize_t aldebaran_get_ecc_info(struct smu_context *smu,
{
struct smu_table_context *smu_table = &smu->smu_table;
EccInfoTable_t *ecc_table = NULL;
+ EccInfoTable_t_v2 *ecc_table_v2 = NULL;
struct ecc_info_per_ch *ecc_info_per_channel = NULL;
int i, ret = 0;
+ int table_version = 0;
struct umc_ecc_info *eccinfo = (struct umc_ecc_info *)table;- ret = aldebaran_check_ecc_table_support(smu);
+ ret = aldebaran_check_ecc_table_support(smu, &table_version);
if (ret)
return ret;- ret = smu_cmn_update_table(smu,
- SMU_TABLE_ECCINFO, - 0, - smu_table->ecc_table, - false);
- if (ret) {
- dev_info(smu->adev->dev, "Failed to export SMU ecc table!\n");
- return ret;
- }
+ if (table_version == 1) {
+ ret = smu_cmn_update_table(smu,
+ SMU_TABLE_ECCINFO, + 0, + smu_table->ecc_table, + false);
+ if (ret) {
+ dev_info(smu->adev->dev, "Failed to export SMU ecc table!\n");
+ return ret;
+ }
+
+ ecc_table = (EccInfoTable_t *)smu_table->ecc_table;
+
+ for (i = 0; i < ALDEBARAN_UMC_CHANNEL_NUM; i++) {
+ ecc_info_per_channel = &(eccinfo->ecc[i]);
+ ecc_info_per_channel->ce_count_lo_chip =
+ ecc_table->EccInfo[i].ce_count_lo_chip;
+ ecc_info_per_channel->ce_count_hi_chip =
+ ecc_table->EccInfo[i].ce_count_hi_chip;
+ ecc_info_per_channel->mca_umc_status =
+ ecc_table->EccInfo[i].mca_umc_status;
+ ecc_info_per_channel->mca_umc_addr =
+ ecc_table->EccInfo[i].mca_umc_addr;
+ }
+ } else if (table_version == 2) {
+ /* still use SMU_TABLE_ECC_INFO index,
+ * smu 68.55.0 add mca_ceumc_addr variable + * in EccInfo_t struct to report correctable + * error address and the table_id is not changed + */ + ret = smu_cmn_update_table(smu, + SMU_TABLE_ECCINFO, + 0, + smu_table->ecc_table_v2,
+ false);
- ecc_table = (EccInfoTable_t *)smu_table->ecc_table;
-
- for (i = 0; i < ALDEBARAN_UMC_CHANNEL_NUM; i++) {
- ecc_info_per_channel = &(eccinfo->ecc[i]);
- ecc_info_per_channel->ce_count_lo_chip =
- ecc_table->EccInfo[i].ce_count_lo_chip;
- ecc_info_per_channel->ce_count_hi_chip =
- ecc_table->EccInfo[i].ce_count_hi_chip;
- ecc_info_per_channel->mca_umc_status =
- ecc_table->EccInfo[i].mca_umc_status;
- ecc_info_per_channel->mca_umc_addr =
- ecc_table->EccInfo[i].mca_umc_addr;
+ if (ret) {
+ dev_info(smu->adev->dev, "Failed to export SMU ecc table_v2!\n");
+ return ret;
+ }
+
+ ecc_table_v2 = (EccInfoTable_t_v2 *)smu_table->ecc_table_v2;
+
+ for (i = 0; i < ALDEBARAN_UMC_CHANNEL_NUM; i++) {
+ ecc_info_per_channel = &(eccinfo->ecc[i]);
+ ecc_info_per_channel->ce_count_lo_chip =
+ ecc_table_v2->EccInfo[i].ce_count_lo_chip;
+ ecc_info_per_channel->ce_count_hi_chip =
+ ecc_table_v2->EccInfo[i].ce_count_hi_chip;
+ ecc_info_per_channel->mca_umc_status =
+ ecc_table_v2->EccInfo[i].mca_umc_status;
+ ecc_info_per_channel->mca_umc_addr =
+ ecc_table_v2->EccInfo[i].mca_umc_addr;
+ ecc_info_per_channel->mca_ceumc_addr =
+ ecc_table_v2->EccInfo[i].mca_ceumc_addr;
+ }
}
return ret;drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c +2 -0
@@ -552,9 +552,11 @@ int smu_v13_0_fini_smc_tables(struct smu_context *smu)kfree (smu_table->hardcode_pptable); smu_table->hardcode_pptable =NULL ; +kfree (smu_table->ecc_table_v2);kfree (smu_table->ecc_table);kfree (smu_table->metrics_table);kfree (smu_table->watermarks_table); + smu_table->ecc_table_v2 =NULL ; smu_table->ecc_table =NULL ; smu_table->metrics_table =NULL ; smu_table->watermarks_table =NULL ;
--
2.17.1
[AMD Official Use Only - General]
Hi Kevin,
发件人: Wang, Yang(Kevin) <[email protected]>
发送时间: Monday, May 23, 2022 4:49 PM
收件人: Yang, Stanley <[email protected]>; [email protected]; Zhang, Hawking <[email protected]>; Zhou1, Tao <[email protected]>; Quan, Evan <[email protected]>; Lazar, Lijo <[email protected]> 主题: Re: [PATCH Review 1/2] drm/amdgpu/pm: support mca_ceumc_addr in ecctable
[AMD Official Use Only - General]
________________________________
From: amd-gfx <[email protected]<mailto:[email protected]>> on behalf of Stanley.Yang <[email protected]<mailto:[email protected]>>
Sent: Monday, May 23, 2022 4:17 PM
To: [email protected]<mailto:[email protected]> <[email protected]<mailto:[email protected]>>; Zhang, Hawking <[email protected]<mailto:[email protected]>>; Zhou1, Tao <[email protected]<mailto:[email protected]>>; Quan, Evan <[email protected]<mailto:[email protected]>>; Lazar, Lijo <[email protected]<mailto:[email protected]>>
Cc: Yang, Stanley <[email protected]<mailto:[email protected]>>
Subject: [PATCH Review 1/2] drm/amdgpu/pm: support mca_ceumc_addr in ecctable
SMU add a new variable mca_ceumc_addr to record umc correctable error address in EccInfo table, driver side add ecctable_v2 to support this feature
--- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h | 1 + drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h | 2 + .../inc/pmfw_if/smu13_driver_if_aldebaran.h | 15 +++ .../drm/amd/pm/swsmu/smu13/aldebaran_ppt.c | 101 ++++++++++++++---- .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c | 2 + 5 files changed, 98 insertions(+), 23 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h +1 -0
@@ -328,6 +328,7 @@ struct ecc_info_per_ch {uint16_t ce_count_hi_chip;uint64_t mca_umc_status;uint64_t mca_umc_addr; +uint64_t mca_ceumc_addr; };struct umc_ecc_info {
drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h +2 -0
@@ -322,6 +322,7 @@ enum smu_table_idSMU_TABLE_PACE ,SMU_TABLE_ECCINFO ,SMU_TABLE_COMBO_PPTABLE , +SMU_TABLE_ECCINFO_V2 , SMU_TABLE_COUNT, };
@@ -340,6 +341,7 @@ struct smu_table_contextvoid *driver_pptable;void *combo_pptable;void *ecc_table; +void *ecc_table_v2;// adapt to smu support record mca_ceumc_addr void *driver_smu_config_table;struct smu_table tables[SMU_TABLE_COUNT ];/*
drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu13_driver_if_aldebaran.h +15 -0
@@ -522,6 +522,21 @@ typedef struct {EccInfo_t EccInfo[ALDEBARAN_UMC_CHANNEL_NUM ]; } EccInfoTable_t; +typedef struct { +uint64_t mca_umc_status; +uint64_t mca_umc_addr; +uint64_t mca_ceumc_addr; + +uint16_t ce_count_lo_chip; +uint16_t ce_count_hi_chip; + +uint32_t eccPadding; +}EccInfo_t_v2 ; + +typedef struct { +EccInfo_t_v2 EccInfo[ALDEBARAN_UMC_CHANNEL_NUM ]; +}EccInfoTable_t_v2 ; +// These defines are used with the following messages: // SMC_MSG_TransferTableDram2Smu // SMC_MSG_TransferTableSmu2Dram
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +9 -0
@@ -82,6 +82,12 @@*/ #define SUPPORT_ECCTABLE_SMU_VERSION 0x00442a00 +/* +* SMU support mca_ceumc_addr in ECCTABLE since version 68.55.0, +* use this to check mca_ceumc_addr record whether support +*/ +#define SUPPORT_ECCTABLE_V2_SMU_VERSION 0x00443700 +/* * SMU support BAD CHENNEL info MSG since version 68.51.00, * use this to check ECCTALE feature whether support
@@ -239,6 +245,9 @@ static int aldebaran_tables_init(struct smu_context *smu)SMU_TABLE_INIT (tables,SMU_TABLE_ECCINFO ,sizeof (EccInfoTable_t),PAGE_SIZE ,AMDGPU_GEM_DOMAIN_VRAM ); +SMU_TABLE_INIT (tables,SMU_TABLE_ECCINFO_V2 ,sizeof (EccInfoTable_t_v2), +PAGE_SIZE ,AMDGPU_GEM_DOMAIN_VRAM ); +
[kevin]:
this table mapping is not needed, the reason as below.
smu_table->metrics_table = kzalloc(sizeof(SmuMetrics_t), GFP_KERNEL);
if (!smu_table->metrics_table)
return -ENOMEM;
@@ -255,6 +264,10 @@ static int aldebaran_tables_init(struct smu_context *smu)
if (!smu_table->ecc_table)
return -ENOMEM;+ smu_table->ecc_table_v2 = kzalloc(tables[SMU_TABLE_ECCINFO_V2].size, GFP_KERNEL);
+ if (!smu_table->ecc_table_v2)
+ return -ENOMEM;;
+
[kevin]:
add eccinfo table v2 is not needed for this case, this table is only used store table data from pmfw to driver, you can create a large enough table which can save ecc table data directly.
e.g:
size = max(sizeof(EccInfoTable_t_v2), sizeof(EccInfoTable_t)); smu_table->ecc_table = kzalloc(size, GFP_KERNEL);
Best Regards,
Kevin
[Yang, Stanley] : this method is not forward compatible, or driver need complex convert to get the correct value, if new driver use an old pmfw.
return 0;
}
@@ -1802,7 +1815,8 @@ static ssize_t aldebaran_get_gpu_metrics(struct smu_context *smu,
return sizeof(struct gpu_metrics_v1_3);
}
-static int aldebaran_check_ecc_table_support(struct smu_context *smu) +static int aldebaran_check_ecc_table_support(struct smu_context *smu,
+ int *ecctable_version)
{
uint32_t if_version = 0xff, smu_version = 0xff;
int ret = 0;@@ -1815,6 +1829,11 @@ static int aldebaran_check_ecc_table_support(struct smu_context *smu)
if (smu_version < SUPPORT_ECCTABLE_SMU_VERSION)
ret = -EOPNOTSUPP;
+ else if (smu_version >= SUPPORT_ECCTABLE_SMU_VERSION &&
+ smu_version < SUPPORT_ECCTABLE_V2_SMU_VERSION)
+ *ecctable_version = 1;
+ else
+ *ecctable_version = 2;
return ret;
}
@@ -1824,36 +1843,72 @@ static ssize_t aldebaran_get_ecc_info(struct smu_context *smu,
{
struct smu_table_context *smu_table = &smu->smu_table;
EccInfoTable_t *ecc_table = NULL;
+ EccInfoTable_t_v2 *ecc_table_v2 = NULL;
struct ecc_info_per_ch *ecc_info_per_channel = NULL;
int i, ret = 0;
+ int table_version = 0;
struct umc_ecc_info *eccinfo = (struct umc_ecc_info *)table;- ret = aldebaran_check_ecc_table_support(smu);
+ ret = aldebaran_check_ecc_table_support(smu, &table_version);
if (ret)
return ret;- ret = smu_cmn_update_table(smu,
- SMU_TABLE_ECCINFO, - 0, - smu_table->ecc_table, - false);
- if (ret) {
- dev_info(smu->adev->dev, "Failed to export SMU ecc table!\n");
- return ret;
- }
+ if (table_version == 1) {
+ ret = smu_cmn_update_table(smu,
+ SMU_TABLE_ECCINFO, + 0, + smu_table->ecc_table, + false);
+ if (ret) {
+ dev_info(smu->adev->dev, "Failed to export SMU ecc table!\n");
+ return ret;
+ }
+
+ ecc_table = (EccInfoTable_t *)smu_table->ecc_table;
+
+ for (i = 0; i < ALDEBARAN_UMC_CHANNEL_NUM; i++) {
+ ecc_info_per_channel = &(eccinfo->ecc[i]);
+ ecc_info_per_channel->ce_count_lo_chip =
+ ecc_table->EccInfo[i].ce_count_lo_chip;
+ ecc_info_per_channel->ce_count_hi_chip =
+ ecc_table->EccInfo[i].ce_count_hi_chip;
+ ecc_info_per_channel->mca_umc_status =
+ ecc_table->EccInfo[i].mca_umc_status;
+ ecc_info_per_channel->mca_umc_addr =
+ ecc_table->EccInfo[i].mca_umc_addr;
+ }
+ } else if (table_version == 2) {
+ /* still use SMU_TABLE_ECC_INFO index,
+ * smu 68.55.0 add mca_ceumc_addr variable + * in EccInfo_t struct to report correctable + * error address and the table_id is not changed + */ + ret = smu_cmn_update_table(smu, + SMU_TABLE_ECCINFO, + 0, + smu_table->ecc_table_v2,
+ false);
- ecc_table = (EccInfoTable_t *)smu_table->ecc_table;
-
- for (i = 0; i < ALDEBARAN_UMC_CHANNEL_NUM; i++) {
- ecc_info_per_channel = &(eccinfo->ecc[i]);
- ecc_info_per_channel->ce_count_lo_chip =
- ecc_table->EccInfo[i].ce_count_lo_chip;
- ecc_info_per_channel->ce_count_hi_chip =
- ecc_table->EccInfo[i].ce_count_hi_chip;
- ecc_info_per_channel->mca_umc_status =
- ecc_table->EccInfo[i].mca_umc_status;
- ecc_info_per_channel->mca_umc_addr =
- ecc_table->EccInfo[i].mca_umc_addr;
+ if (ret) {
+ dev_info(smu->adev->dev, "Failed to export SMU ecc table_v2!\n");
+ return ret;
+ }
+
+ ecc_table_v2 = (EccInfoTable_t_v2 *)smu_table->ecc_table_v2;
+
+ for (i = 0; i < ALDEBARAN_UMC_CHANNEL_NUM; i++) {
+ ecc_info_per_channel = &(eccinfo->ecc[i]);
+ ecc_info_per_channel->ce_count_lo_chip =
+ ecc_table_v2->EccInfo[i].ce_count_lo_chip;
+ ecc_info_per_channel->ce_count_hi_chip =
+ ecc_table_v2->EccInfo[i].ce_count_hi_chip;
+ ecc_info_per_channel->mca_umc_status =
+ ecc_table_v2->EccInfo[i].mca_umc_status;
+ ecc_info_per_channel->mca_umc_addr =
+ ecc_table_v2->EccInfo[i].mca_umc_addr;
+ ecc_info_per_channel->mca_ceumc_addr =
+ ecc_table_v2->EccInfo[i].mca_ceumc_addr;
+ }
}
return ret;drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c +2 -0
@@ -552,9 +552,11 @@ int smu_v13_0_fini_smc_tables(struct smu_context *smu)kfree (smu_table->hardcode_pptable); smu_table->hardcode_pptable =NULL ; +kfree (smu_table->ecc_table_v2);kfree (smu_table->ecc_table);kfree (smu_table->metrics_table);kfree (smu_table->watermarks_table); + smu_table->ecc_table_v2 =NULL ; smu_table->ecc_table =NULL ; smu_table->metrics_table =NULL ; smu_table->watermarks_table =NULL ;
--
2.17.1
[AMD Official Use Only - General]
[AMD Official Use Only - General]
Hi Kevin,
Please ignore above mail, thanks your suggestion, I will try it.
Regards,
Stanley
发件人: amd-gfx <[email protected]> 代表 Yang, Stanley <[email protected]>
日期: 星期一, 2022年5月23日 下午6:16
收件人: Wang, Yang(Kevin) <[email protected]>, [email protected] <[email protected]>, Zhang, Hawking <[email protected]>, Zhou1, Tao <[email protected]>, Quan, Evan <[email protected]>, Lazar, Lijo <[email protected]> 主题: 回复: [PATCH Review 1/2] drm/amdgpu/pm: support mca_ceumc_addr in ecctable
[AMD Official Use Only - General]
[AMD Official Use Only - General]
Hi Kevin,
发件人: Wang, Yang(Kevin) <[email protected]>
发送时间: Monday, May 23, 2022 4:49 PM
收件人: Yang, Stanley <[email protected]>; [email protected]; Zhang, Hawking <[email protected]>; Zhou1, Tao <[email protected]>; Quan, Evan <[email protected]>; Lazar, Lijo <[email protected]> 主题: Re: [PATCH Review 1/2] drm/amdgpu/pm: support mca_ceumc_addr in ecctable
[AMD Official Use Only - General]
________________________________
From: amd-gfx <[email protected]<mailto:[email protected]>> on behalf of Stanley.Yang <[email protected]<mailto:[email protected]>>
Sent: Monday, May 23, 2022 4:17 PM
To: [email protected]<mailto:[email protected]> <[email protected]<mailto:[email protected]>>; Zhang, Hawking <[email protected]<mailto:[email protected]>>; Zhou1, Tao <[email protected]<mailto:[email protected]>>; Quan, Evan <[email protected]<mailto:[email protected]>>; Lazar, Lijo <[email protected]<mailto:[email protected]>>
Cc: Yang, Stanley <[email protected]<mailto:[email protected]>>
Subject: [PATCH Review 1/2] drm/amdgpu/pm: support mca_ceumc_addr in ecctable
SMU add a new variable mca_ceumc_addr to record umc correctable error address in EccInfo table, driver side add ecctable_v2 to support this feature
--- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h | 1 + drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h | 2 + .../inc/pmfw_if/smu13_driver_if_aldebaran.h | 15 +++ .../drm/amd/pm/swsmu/smu13/aldebaran_ppt.c | 101 ++++++++++++++---- .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c | 2 + 5 files changed, 98 insertions(+), 23 deletions(-)
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h +1 -0
@@ -328,6 +328,7 @@ struct ecc_info_per_ch {uint16_t ce_count_hi_chip;uint64_t mca_umc_status;uint64_t mca_umc_addr; +uint64_t mca_ceumc_addr; };struct umc_ecc_info {
drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h +2 -0
@@ -322,6 +322,7 @@ enum smu_table_idSMU_TABLE_PACE ,SMU_TABLE_ECCINFO ,SMU_TABLE_COMBO_PPTABLE , +SMU_TABLE_ECCINFO_V2 , SMU_TABLE_COUNT, };
@@ -340,6 +341,7 @@ struct smu_table_contextvoid *driver_pptable;void *combo_pptable;void *ecc_table; +void *ecc_table_v2;// adapt to smu support record mca_ceumc_addr void *driver_smu_config_table;struct smu_table tables[SMU_TABLE_COUNT ];/*
drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu13_driver_if_aldebaran.h +15 -0
@@ -522,6 +522,21 @@ typedef struct {EccInfo_t EccInfo[ALDEBARAN_UMC_CHANNEL_NUM ]; } EccInfoTable_t; +typedef struct { +uint64_t mca_umc_status; +uint64_t mca_umc_addr; +uint64_t mca_ceumc_addr; + +uint16_t ce_count_lo_chip; +uint16_t ce_count_hi_chip; + +uint32_t eccPadding; +}EccInfo_t_v2 ; + +typedef struct { +EccInfo_t_v2 EccInfo[ALDEBARAN_UMC_CHANNEL_NUM ]; +}EccInfoTable_t_v2 ; +// These defines are used with the following messages: // SMC_MSG_TransferTableDram2Smu // SMC_MSG_TransferTableSmu2Dram
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +9 -0
@@ -82,6 +82,12 @@*/ #define SUPPORT_ECCTABLE_SMU_VERSION 0x00442a00 +/* +* SMU support mca_ceumc_addr in ECCTABLE since version 68.55.0, +* use this to check mca_ceumc_addr record whether support +*/ +#define SUPPORT_ECCTABLE_V2_SMU_VERSION 0x00443700 +/* * SMU support BAD CHENNEL info MSG since version 68.51.00, * use this to check ECCTALE feature whether support
@@ -239,6 +245,9 @@ static int aldebaran_tables_init(struct smu_context *smu)SMU_TABLE_INIT (tables,SMU_TABLE_ECCINFO ,sizeof (EccInfoTable_t),PAGE_SIZE ,AMDGPU_GEM_DOMAIN_VRAM ); +SMU_TABLE_INIT (tables,SMU_TABLE_ECCINFO_V2 ,sizeof (EccInfoTable_t_v2), +PAGE_SIZE ,AMDGPU_GEM_DOMAIN_VRAM ); +
[kevin]:
this table mapping is not needed, the reason as below.
smu_table->metrics_table = kzalloc(sizeof(SmuMetrics_t), GFP_KERNEL);
if (!smu_table->metrics_table)
return -ENOMEM;
@@ -255,6 +264,10 @@ static int aldebaran_tables_init(struct smu_context *smu)
if (!smu_table->ecc_table)
return -ENOMEM;+ smu_table->ecc_table_v2 = kzalloc(tables[SMU_TABLE_ECCINFO_V2].size, GFP_KERNEL);
+ if (!smu_table->ecc_table_v2)
+ return -ENOMEM;;
+
[kevin]:
add eccinfo table v2 is not needed for this case, this table is only used store table data from pmfw to driver, you can create a large enough table which can save ecc table data directly.
e.g:
size = max(sizeof(EccInfoTable_t_v2), sizeof(EccInfoTable_t)); smu_table->ecc_table = kzalloc(size, GFP_KERNEL);
Best Regards,
Kevin
[Yang, Stanley] : this method is not forward compatible, or driver need complex convert to get the correct value, if new driver use an old pmfw.
return 0;
}
@@ -1802,7 +1815,8 @@ static ssize_t aldebaran_get_gpu_metrics(struct smu_context *smu,
return sizeof(struct gpu_metrics_v1_3);
}
-static int aldebaran_check_ecc_table_support(struct smu_context *smu) +static int aldebaran_check_ecc_table_support(struct smu_context *smu,
+ int *ecctable_version)
{
uint32_t if_version = 0xff, smu_version = 0xff;
int ret = 0;@@ -1815,6 +1829,11 @@ static int aldebaran_check_ecc_table_support(struct smu_context *smu)
if (smu_version < SUPPORT_ECCTABLE_SMU_VERSION)
ret = -EOPNOTSUPP;
+ else if (smu_version >= SUPPORT_ECCTABLE_SMU_VERSION &&
+ smu_version < SUPPORT_ECCTABLE_V2_SMU_VERSION)
+ *ecctable_version = 1;
+ else
+ *ecctable_version = 2;
return ret;
}
@@ -1824,36 +1843,72 @@ static ssize_t aldebaran_get_ecc_info(struct smu_context *smu,
{
struct smu_table_context *smu_table = &smu->smu_table;
EccInfoTable_t *ecc_table = NULL;
+ EccInfoTable_t_v2 *ecc_table_v2 = NULL;
struct ecc_info_per_ch *ecc_info_per_channel = NULL;
int i, ret = 0;
+ int table_version = 0;
struct umc_ecc_info *eccinfo = (struct umc_ecc_info *)table;- ret = aldebaran_check_ecc_table_support(smu);
+ ret = aldebaran_check_ecc_table_support(smu, &table_version);
if (ret)
return ret;- ret = smu_cmn_update_table(smu,
- SMU_TABLE_ECCINFO, - 0, - smu_table->ecc_table, - false);
- if (ret) {
- dev_info(smu->adev->dev, "Failed to export SMU ecc table!\n");
- return ret;
- }
+ if (table_version == 1) {
+ ret = smu_cmn_update_table(smu,
+ SMU_TABLE_ECCINFO, + 0, + smu_table->ecc_table, + false);
+ if (ret) {
+ dev_info(smu->adev->dev, "Failed to export SMU ecc table!\n");
+ return ret;
+ }
+
+ ecc_table = (EccInfoTable_t *)smu_table->ecc_table;
+
+ for (i = 0; i < ALDEBARAN_UMC_CHANNEL_NUM; i++) {
+ ecc_info_per_channel = &(eccinfo->ecc[i]);
+ ecc_info_per_channel->ce_count_lo_chip =
+ ecc_table->EccInfo[i].ce_count_lo_chip;
+ ecc_info_per_channel->ce_count_hi_chip =
+ ecc_table->EccInfo[i].ce_count_hi_chip;
+ ecc_info_per_channel->mca_umc_status =
+ ecc_table->EccInfo[i].mca_umc_status;
+ ecc_info_per_channel->mca_umc_addr =
+ ecc_table->EccInfo[i].mca_umc_addr;
+ }
+ } else if (table_version == 2) {
+ /* still use SMU_TABLE_ECC_INFO index,
+ * smu 68.55.0 add mca_ceumc_addr variable + * in EccInfo_t struct to report correctable + * error address and the table_id is not changed + */ + ret = smu_cmn_update_table(smu, + SMU_TABLE_ECCINFO, + 0, + smu_table->ecc_table_v2,
+ false);
- ecc_table = (EccInfoTable_t *)smu_table->ecc_table;
-
- for (i = 0; i < ALDEBARAN_UMC_CHANNEL_NUM; i++) {
- ecc_info_per_channel = &(eccinfo->ecc[i]);
- ecc_info_per_channel->ce_count_lo_chip =
- ecc_table->EccInfo[i].ce_count_lo_chip;
- ecc_info_per_channel->ce_count_hi_chip =
- ecc_table->EccInfo[i].ce_count_hi_chip;
- ecc_info_per_channel->mca_umc_status =
- ecc_table->EccInfo[i].mca_umc_status;
- ecc_info_per_channel->mca_umc_addr =
- ecc_table->EccInfo[i].mca_umc_addr;
+ if (ret) {
+ dev_info(smu->adev->dev, "Failed to export SMU ecc table_v2!\n");
+ return ret;
+ }
+
+ ecc_table_v2 = (EccInfoTable_t_v2 *)smu_table->ecc_table_v2;
+
+ for (i = 0; i < ALDEBARAN_UMC_CHANNEL_NUM; i++) {
+ ecc_info_per_channel = &(eccinfo->ecc[i]);
+ ecc_info_per_channel->ce_count_lo_chip =
+ ecc_table_v2->EccInfo[i].ce_count_lo_chip;
+ ecc_info_per_channel->ce_count_hi_chip =
+ ecc_table_v2->EccInfo[i].ce_count_hi_chip;
+ ecc_info_per_channel->mca_umc_status =
+ ecc_table_v2->EccInfo[i].mca_umc_status;
+ ecc_info_per_channel->mca_umc_addr =
+ ecc_table_v2->EccInfo[i].mca_umc_addr;
+ ecc_info_per_channel->mca_ceumc_addr =
+ ecc_table_v2->EccInfo[i].mca_ceumc_addr;
+ }
}
return ret;drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c +2 -0
@@ -552,9 +552,11 @@ int smu_v13_0_fini_smc_tables(struct smu_context *smu)kfree (smu_table->hardcode_pptable); smu_table->hardcode_pptable =NULL ; +kfree (smu_table->ecc_table_v2);kfree (smu_table->ecc_table);kfree (smu_table->metrics_table);kfree (smu_table->watermarks_table); + smu_table->ecc_table_v2 =NULL ; smu_table->ecc_table =NULL ; smu_table->metrics_table =NULL ; smu_table->watermarks_table =NULL ;
--
2.17.1
Quoted 30 lines · Stanley.Yang
SMU add a new variable
mca_ceumc_addrto record umc correctable error address in EccInfo table, driver side addecctable_v2to support this featureSigned-off-by Stanley.Yang <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h | 1 + drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h | 2 + .../inc/pmfw_if/smu13_driver_if_aldebaran.h | 15 +++ .../drm/amd/pm/swsmu/smu13/aldebaran_ppt.c | 101 ++++++++++++++---- .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c | 2 + 5 files changed, 98 insertions(+), 23 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h +1
-0@@ -328,6 +328,7 @@ struct ecc_info_per_ch {uint16_t ce_count_hi_chip;uint64_t mca_umc_status;uint64_t mca_umc_addr; +uint64_t mca_ceumc_addr; };struct umc_ecc_info {drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h +0
-0
drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h enum smu_table_id · line 325
Quoted 4 lines
drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h +1
-0@@ -322,6 +322,7 @@ enum smu_table_idSMU_TABLE_PACE ,SMU_TABLE_ECCINFO ,SMU_TABLE_COMBO_PPTABLE , + SMU_TABLE_ECCINFO_V2,
Hi Stanley,
This is not the right approach. Need to ask FW team to fix this. There shouldn't be any new id with each version of table. You may check Sienna Cichlid smu metrics table as an example and ask FW team to follow something similar. I don't see 68.55 being released, so it's not late anyway. We don't need to keep defining pointers in table context per version of ECC table.
Thanks,
Lijo
Quoted 209 lines
drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h +1
-0SMU_TABLE_COUNT, };@@ -340,6 +341,7 @@ struct smu_table_contextvoid *driver_pptable;void *combo_pptable;void *ecc_table; +void *ecc_table_v2;// adapt to smu support record mca_ceumc_addr void *driver_smu_config_table;struct smu_table tables[SMU_TABLE_COUNT ];/* drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu13_driver_if_aldebaran.h +15
-0@@ -522,6 +522,21 @@ typedef struct {EccInfo_t EccInfo[ALDEBARAN_UMC_CHANNEL_NUM ]; } EccInfoTable_t; +typedef struct { +uint64_t mca_umc_status; +uint64_t mca_umc_addr; +uint64_t mca_ceumc_addr; + +uint16_t ce_count_lo_chip; +uint16_t ce_count_hi_chip; + +uint32_t eccPadding; +}EccInfo_t_v2 ; + +typedef struct { +EccInfo_t_v2 EccInfo[ALDEBARAN_UMC_CHANNEL_NUM ]; +}EccInfoTable_t_v2 ; +// These defines are used with the following messages: // SMC_MSG_TransferTableDram2Smu // SMC_MSG_TransferTableSmu2Dram drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +78
-23@@ -82,6 +82,12 @@*/ #define SUPPORT_ECCTABLE_SMU_VERSION 0x00442a00 +/* +* SMU support mca_ceumc_addr in ECCTABLE since version 68.55.0, +* use this to check mca_ceumc_addr record whether support +*/ +#define SUPPORT_ECCTABLE_V2_SMU_VERSION 0x00443700 +/* * SMU support BAD CHENNEL info MSG since version 68.51.00, * use this to check ECCTALE feature whether support @@ -239,6 +245,9 @@ static int aldebaran_tables_init(struct smu_context *smu)SMU_TABLE_INIT (tables,SMU_TABLE_ECCINFO ,sizeof (EccInfoTable_t),PAGE_SIZE ,AMDGPU_GEM_DOMAIN_VRAM ); +SMU_TABLE_INIT (tables,SMU_TABLE_ECCINFO_V2 ,sizeof (EccInfoTable_t_v2), +PAGE_SIZE ,AMDGPU_GEM_DOMAIN_VRAM ); + smu_table->metrics_table =kzalloc (sizeof (SmuMetrics_t),GFP_KERNEL );if (!smu_table->metrics_table)return -ENOMEM ;@@ -255,6 +264,10 @@ static int aldebaran_tables_init(struct smu_context *smu)if (!smu_table->ecc_table)return -ENOMEM ; + smu_table->ecc_table_v2 =kzalloc (tables[SMU_TABLE_ECCINFO_V2 ].size,GFP_KERNEL ); +if (!smu_table->ecc_table_v2) +return -ENOMEM ;; +return 0 ; }@@ -1802,7 +1815,8 @@ static ssize_t aldebaran_get_gpu_metrics(struct smu_context *smu,return sizeof (struct gpu_metrics_v1_3 ); } -static int aldebaran_check_ecc_table_support (struct smu_context *smu) +static int aldebaran_check_ecc_table_support (struct smu_context *smu, +int *ecctable_version) {uint32_t if_version =0xff , smu_version =0xff ;int ret =0 ;@@ -1815,6 +1829,11 @@ static int aldebaran_check_ecc_table_support(struct smu_context *smu)if (smu_version <SUPPORT_ECCTABLE_SMU_VERSION ) ret = -EOPNOTSUPP ; +else if (smu_version >=SUPPORT_ECCTABLE_SMU_VERSION && + smu_version <SUPPORT_ECCTABLE_V2_SMU_VERSION ) + *ecctable_version =1 ; +else + *ecctable_version =2 ;return ret; }@@ -1824,36 +1843,72 @@ static ssize_t aldebaran_get_ecc_info(struct smu_context *smu, {struct smu_table_context *smu_table = &smu->smu_table;EccInfoTable_t *ecc_table =NULL ; +EccInfoTable_t_v2 *ecc_table_v2 =NULL ;struct ecc_info_per_ch *ecc_info_per_channel =NULL ;int i, ret =0 ; +int table_version =0 ;struct umc_ecc_info *eccinfo = (struct umc_ecc_info *)table; - ret =aldebaran_check_ecc_table_support (smu); + ret =aldebaran_check_ecc_table_support (smu, &table_version);if (ret)return ret; - ret =smu_cmn_update_table (smu, -SMU_TABLE_ECCINFO , -0 , - smu_table->ecc_table, - false); -if (ret) { -dev_info (smu->adev->dev,"Failed to export SMU ecc table!\n" ); -return ret; - } +if (table_version ==1 ) { + ret =smu_cmn_update_table (smu, +SMU_TABLE_ECCINFO , +0 , + smu_table->ecc_table, + false); +if (ret) { +dev_info (smu->adev->dev,"Failed to export SMU ecc table!\n" ); +return ret; + } + + ecc_table = (EccInfoTable_t *)smu_table->ecc_table; + +for (i =0 ; i <ALDEBARAN_UMC_CHANNEL_NUM ; i++) { + ecc_info_per_channel = &(eccinfo->ecc[i]); + ecc_info_per_channel->ce_count_lo_chip = + ecc_table->EccInfo[i].ce_count_lo_chip; + ecc_info_per_channel->ce_count_hi_chip = + ecc_table->EccInfo[i].ce_count_hi_chip; + ecc_info_per_channel->mca_umc_status = + ecc_table->EccInfo[i].mca_umc_status; + ecc_info_per_channel->mca_umc_addr = + ecc_table->EccInfo[i].mca_umc_addr; + } + }else if (table_version ==2 ) { +/* still use SMU_TABLE_ECC_INFO index, +* smu 68.55.0 add mca_ceumc_addr variable +* in EccInfo_t struct to report correctable +* error address and the table_id is not changed +*/ + ret =smu_cmn_update_table (smu, +SMU_TABLE_ECCINFO , +0 , + smu_table->ecc_table_v2, + false); - ecc_table = (EccInfoTable_t *)smu_table->ecc_table; - -for (i =0 ; i <ALDEBARAN_UMC_CHANNEL_NUM ; i++) { - ecc_info_per_channel = &(eccinfo->ecc[i]); - ecc_info_per_channel->ce_count_lo_chip = - ecc_table->EccInfo[i].ce_count_lo_chip; - ecc_info_per_channel->ce_count_hi_chip = - ecc_table->EccInfo[i].ce_count_hi_chip; - ecc_info_per_channel->mca_umc_status = - ecc_table->EccInfo[i].mca_umc_status; - ecc_info_per_channel->mca_umc_addr = - ecc_table->EccInfo[i].mca_umc_addr; +if (ret) { +dev_info (smu->adev->dev,"Failed to export SMU ecc table_v2!\n" ); +return ret; + } + + ecc_table_v2 = (EccInfoTable_t_v2 *)smu_table->ecc_table_v2; + +for (i =0 ; i <ALDEBARAN_UMC_CHANNEL_NUM ; i++) { + ecc_info_per_channel = &(eccinfo->ecc[i]); + ecc_info_per_channel->ce_count_lo_chip = + ecc_table_v2->EccInfo[i].ce_count_lo_chip; + ecc_info_per_channel->ce_count_hi_chip = + ecc_table_v2->EccInfo[i].ce_count_hi_chip; + ecc_info_per_channel->mca_umc_status = + ecc_table_v2->EccInfo[i].mca_umc_status; + ecc_info_per_channel->mca_umc_addr = + ecc_table_v2->EccInfo[i].mca_umc_addr; + ecc_info_per_channel->mca_ceumc_addr = + ecc_table_v2->EccInfo[i].mca_ceumc_addr; + } }return ret;drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c +2
-0@@ -552,9 +552,11 @@ int smu_v13_0_fini_smc_tables(struct smu_context *smu)kfree (smu_table->hardcode_pptable); smu_table->hardcode_pptable =NULL ; +kfree (smu_table->ecc_table_v2);kfree (smu_table->ecc_table);kfree (smu_table->metrics_table);kfree (smu_table->watermarks_table); + smu_table->ecc_table_v2 =NULL ; smu_table->ecc_table =NULL ; smu_table->metrics_table =NULL ; smu_table->watermarks_table =NULL ;
[AMD Official Use Only - General]
Hi Lijo,
+@Joo, Maria
Quoted 51 lines
-----邮件原件-----
发件人: Lazar, Lijo <[email protected]>
发送时间: Monday, May 23, 2022 5:12 PM
收件人: Yang, Stanley <[email protected]>; amd- [email protected]; Zhang, Hawking <[email protected]>; Zhou1, Tao <[email protected]>; Quan, Evan <[email protected]>
主题: Re: [PATCH Review 1/2] drm/amdgpu/pm: supportmca_ceumc_addrin ecctableQuoted 32 lines · Stanley.Yang
SMU add a new variable
mca_ceumc_addrto record umc correctable error address in EccInfo table, driver side addecctable_v2to support this featureSigned-off-by Stanley.Yang <[email protected]>--- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h | 1 + drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h | 2 + .../inc/pmfw_if/smu13_driver_if_aldebaran.h | 15 +++ .../drm/amd/pm/swsmu/smu13/aldebaran_ppt.c | 101 ++++++++++++++---- .../gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c | 2 + 5 files changed, 98 insertions(+), 23 deletions(-)drivers/gpu/drm/amd/amdgpu/amdgpu_ras.h +1
-0@@ -328,6 +328,7 @@ struct ecc_info_per_ch {uint16_t ce_count_hi_chip;uint64_t mca_umc_status;uint64_t mca_umc_addr; +uint64_t mca_ceumc_addr; };struct umc_ecc_info {drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h +0
-0drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h enum smu_table_id · line 325
Quoted 4 lines
drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h +1
-0@@ -322,6 +322,7 @@ enum smu_table_idSMU_TABLE_PACE ,SMU_TABLE_ECCINFO ,SMU_TABLE_COMBO_PPTABLE , + SMU_TABLE_ECCINFO_V2,Hi Stanley,
Quoted 8 lines
This is not the right approach. Need to ask FW team to fix this. There shouldn't be any new id with each version of table. You may check Sienna Cichlid smu metrics table as an example and ask FW team to follow something similar. I don't see 68.55 being released, so it's not late anyway. We don't need to keep defining pointers in table context per version of ECC table.
Thanks,
Lijo
[Yang, Stanley] : There is not enough padding space in ecc_table, you can check below struct,
the new added variable is uint32_t type, I think smu can't add uint32_t type in ecc_table directly without change ecc_tabe size. If you have any better approach, we can discuss a better method to complete it.
512 typedef struct {
513 uint64_t mca_umc_status; 514 uint64_t mca_umc_addr; 515 uint16_t ce_count_lo_chip; 516 uint16_t ce_count_hi_chip;
517
518 uint32_t eccPadding;
519 } EccInfo_t;
Thanks,
Stanley
Quoted 230 lines
drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h struct smu_table_context · line 344
Quoted 9 lines
drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h +1
-0SMU_TABLE_COUNT, };@@ -340,6 +341,7 @@ struct smu_table_contextvoid *driver_pptable;void *combo_pptable;void *ecc_table; +void *ecc_table_v2;// adapt to smu support record
mca_ceumc_addrQuoted 5 lines
drivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h +0
-0void *driver_smu_config_table;struct smu_table tables[SMU_TABLE_COUNT ];/* diff --git
a/drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu13_driver_if_aldebaran.hQuoted 1 line
b/drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu13_driver_if_aldebaran.hQuoted 3 lines
index
0f67c56c2863..2868604eff49100644
---
a/drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu13_driver_if_aldebaran.hQuoted 1 line
+++
b/drivers/gpu/drm/amd/pm/swsmu/inc/pmfw_if/smu13_driver_if_aldebar
Quoted 44 lines
+++
an.hdrivers/gpu/drm/amd/pm/swsmu/inc/amdgpu_smu.h +15
-0@@ -522,6 +522,21 @@ typedef struct {EccInfo_t EccInfo[ALDEBARAN_UMC_CHANNEL_NUM ]; } EccInfoTable_t; +typedef struct { +uint64_t mca_umc_status; +uint64_t mca_umc_addr; +uint64_t mca_ceumc_addr; + +uint16_t ce_count_lo_chip; +uint16_t ce_count_hi_chip; + +uint32_t eccPadding; +}EccInfo_t_v2 ; + +typedef struct { +EccInfo_t_v2 EccInfo[ALDEBARAN_UMC_CHANNEL_NUM ]; +}EccInfoTable_t_v2 ; +// These defines are used with the following messages: // SMC_MSG_TransferTableDram2Smu // SMC_MSG_TransferTableSmu2Dram drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +6
-0@@ -82,6 +82,12 @@*/ #define SUPPORT_ECCTABLE_SMU_VERSION 0x00442a00 +/* +* SMU support mca_ceumc_addr in ECCTABLE since version 68.55.0, +* use this to check mca_ceumc_addr record whether support */ +#define SUPPORT_ECCTABLE_V2_SMU_VERSION 0x00443700 +/* * SMU support BAD CHENNEL info MSG since version 68.51.00, * use this to check ECCTALE feature whether support @@ -239,6 +245,9 @@ static int aldebaran_tables_init(struct smu_context *smu) SMU_TABLE_INIT(tables,
SMU_TABLE_ECCINFO,sizeof(EccInfoTable_t),
PAGE_SIZE,AMDGPU_GEM_DOMAIN_VRAM);Quoted 1 line
+ SMU_TABLE_INIT(tables,
SMU_TABLE_ECCINFO_V2,
sizeof(EccInfoTable_t_v2),drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c static int aldebaran_tables_init(struct smu_context
Quoted 6 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +2
-0+ PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM); + smu_table->metrics_table =kzalloc (sizeof (SmuMetrics_t),GFP_KERNEL );if (!smu_table->metrics_table)return -ENOMEM ;@@ -255,6 +264,10 @@ static int aldebaran_tables_init(struct smu_context*smu)
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c static int aldebaran_tables_init(struct smu_context *smu)
Quoted 4 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +1
-0if (!smu_table->ecc_table)return -ENOMEM ; + smu_table->ecc_table_v2 =
kzalloc(tables[SMU_TABLE_ECCINFO_V2].size, GFP_KERNEL);drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c static ssize_t aldebaran_get_gpu_metrics(struct
Quoted 7 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +3
-0+if (!smu_table->ecc_table_v2) +return -ENOMEM ;; +return 0 ; }@@ -1802,7 +1815,8 @@ static ssize_t aldebaran_get_gpu_metrics(struct
smu_context*smu,Quoted 10 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +2
-1return sizeof (struct gpu_metrics_v1_3 ); } -static int aldebaran_check_ecc_table_support (struct smu_context *smu) +static int aldebaran_check_ecc_table_support (struct smu_context *smu, +int *ecctable_version) {uint32_t if_version =0xff , smu_version =0xff ;int ret =0 ;drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c static int · line 1834
Quoted 6 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +2
-0@@ -1815,6 +1829,11 @@ static int aldebaran_check_ecc_table_support(struct smu_context *smu) if (smu_version < SUPPORT_ECCTABLE_SMU_VERSION) ret = -EOPNOTSUPP ; + else if (smu_version >= SUPPORT_ECCTABLE_SMU_VERSION && + smu_version <
SUPPORT_ECCTABLE_V2_SMU_VERSION)drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c static ssize_t aldebaran_get_ecc_info(struct
Quoted 7 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +3
-0+ *ecctable_version =1 ; + else + *ecctable_version =2 ;return ret;}drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +0
-0@@ -1824,36 +1843,72 @@ static ssize_t aldebaran_get_ecc_info(struct
smu_context*smu,Quoted 21 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +3
-8{struct smu_table_context *smu_table = &smu->smu_table;EccInfoTable_t *ecc_table =NULL ; +EccInfoTable_t_v2 *ecc_table_v2 =NULL ;struct ecc_info_per_ch *ecc_info_per_channel =NULL ;int i, ret =0 ; +int table_version =0 ;struct umc_ecc_info *eccinfo = (struct umc_ecc_info *)table; - ret =aldebaran_check_ecc_table_support (smu); + ret =aldebaran_check_ecc_table_support (smu, &table_version);if (ret)return ret; - ret =smu_cmn_update_table (smu, -SMU_TABLE_ECCINFO , -0 , - smu_table->ecc_table, - false); -if (ret) { -dev_info (smu->adev->dev,"Failed to export SMU ecc table!\n" );drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c static ssize_t aldebaran_get_ecc_info(struct smu_context *smu,
Quoted 10 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +8
-2-return ret; - } +if (table_version ==1 ) { + ret =smu_cmn_update_table (smu, +SMU_TABLE_ECCINFO , +0 , + smu_table->ecc_table, + false); +if (ret) { + dev_info(smu->adev->dev, "Failed to export SMU ecctable!\n");
Quoted 32 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +27
-4+return ret; + } + + ecc_table = (EccInfoTable_t *)smu_table->ecc_table; + +for (i =0 ; i <ALDEBARAN_UMC_CHANNEL_NUM ; i++) { + ecc_info_per_channel = &(eccinfo->ecc[i]); + ecc_info_per_channel->ce_count_lo_chip = + ecc_table->EccInfo[i].ce_count_lo_chip; + ecc_info_per_channel->ce_count_hi_chip = + ecc_table->EccInfo[i].ce_count_hi_chip; + ecc_info_per_channel->mca_umc_status = + ecc_table->EccInfo[i].mca_umc_status; + ecc_info_per_channel->mca_umc_addr = + ecc_table->EccInfo[i].mca_umc_addr; + } + }else if (table_version ==2 ) { +/* still use SMU_TABLE_ECC_INFO index, +* smu 68.55.0 add mca_ceumc_addr variable +* in EccInfo_t struct to report correctable +* error address and the table_id is not changed +*/ + ret =smu_cmn_update_table (smu, +SMU_TABLE_ECCINFO , +0 , + smu_table->ecc_table_v2, + false); - ecc_table = (EccInfoTable_t *)smu_table->ecc_table; - -for (i =0 ; i <ALDEBARAN_UMC_CHANNEL_NUM ; i++) { - ecc_info_per_channel = &(eccinfo->ecc[i]);drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c static ssize_t aldebaran_get_ecc_info(struct smu_context *smu,
Quoted 10 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +2
-8- ecc_info_per_channel->ce_count_lo_chip = - ecc_table->EccInfo[i].ce_count_lo_chip; - ecc_info_per_channel->ce_count_hi_chip = - ecc_table->EccInfo[i].ce_count_hi_chip; - ecc_info_per_channel->mca_umc_status = - ecc_table->EccInfo[i].mca_umc_status; - ecc_info_per_channel->mca_umc_addr = - ecc_table->EccInfo[i].mca_umc_addr; +if (ret) { + dev_info(smu->adev->dev, "Failed to export SMU ecctable_v2!\n");
drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c int smu_v13_0_fini_smc_tables(struct smu_context
Quoted 27 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/aldebaran_ppt.c +18
-0+return ret; + } + + ecc_table_v2 = (EccInfoTable_t_v2 *)smu_table->ecc_table_v2; + +for (i =0 ; i <ALDEBARAN_UMC_CHANNEL_NUM ; i++) { + ecc_info_per_channel = &(eccinfo->ecc[i]); + ecc_info_per_channel->ce_count_lo_chip = + ecc_table_v2->EccInfo[i].ce_count_lo_chip; + ecc_info_per_channel->ce_count_hi_chip = + ecc_table_v2->EccInfo[i].ce_count_hi_chip; + ecc_info_per_channel->mca_umc_status = + ecc_table_v2->EccInfo[i].mca_umc_status; + ecc_info_per_channel->mca_umc_addr = + ecc_table_v2->EccInfo[i].mca_umc_addr; + ecc_info_per_channel->mca_ceumc_addr = + ecc_table_v2->EccInfo[i].mca_ceumc_addr; + } }return ret;drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c +0
-0@@ -552,9 +552,11 @@ int smu_v13_0_fini_smc_tables(struct smu_context*smu)
Quoted 12 lines
drivers/gpu/drm/amd/pm/swsmu/smu13/smu_v13_0.c +2
-0kfree (smu_table->hardcode_pptable); smu_table->hardcode_pptable =NULL ; +kfree (smu_table->ecc_table_v2);kfree (smu_table->ecc_table);kfree (smu_table->metrics_table);kfree (smu_table->watermarks_table); + smu_table->ecc_table_v2 =NULL ; smu_table->ecc_table =NULL ; smu_table->metrics_table =NULL ; smu_table->watermarks_table =NULL ;