x86/MCE/AMD, EDAC/mce_amd: Add new SMCA bank types.
Subsystems (6)
PATCH 1/3 x86/MCE/AMD, EDAC/mce_amd: Add new SMCA bank types.
From: Muralidhara M K <[email protected]>
Add the (HWID, MCATYPE) tuples and names for new
SMCA bank types.
Also, add their respective error descriptions to the MCE decoding module edac_mce_amd.
--- arch/x86/include/asm/mce.h | 5 +++ arch/x86/kernel/cpu/mce/amd.c | 16 ++++++++ drivers/edac/mce_amd.c | 70 +++++++++++++++++++++++++++++++++++ 3 files changed, 91 insertions(+)
arch/x86/include/asm/mce.h +5 -0
@@ -317,6 +317,7 @@ enum smca_bank_types {SMCA_CS_V2 ,/* Coherent Slave */ SMCA_PIE ,/* Power, Interrupts, etc. */ SMCA_UMC ,/* Unified Memory Controller */ +SMCA_UMC_V2 ,/* Unified Memory Controller */ SMCA_PB ,/* Parameter Block */ SMCA_PSP ,/* Platform Security Processor */ SMCA_PSP_V2, /* Platform Security Processor */
@@ -325,6 +326,10 @@ enum smca_bank_types {SMCA_MP5 ,/* Microprocessor 5 Unit */ SMCA_NBIO ,/* Northbridge IO Unit */ SMCA_PCIE ,/* PCI Express Unit */ +SMCA_PCIE_V2 ,/* PCI Express Unit */ +SMCA_XGMI_PCS ,/* xGMI PCS Unit */ +SMCA_XGMI_PHY ,/* xGMI PHY Unit */ +SMCA_WAFL_PHY ,/* WAFL PHY Unit */ N_SMCA_BANK_TYPES };
arch/x86/kernel/cpu/mce/amd.c +16 -0
@@ -90,6 +90,7 @@ static struct smca_bank_name smca_names[] = { [SMCA_CS_V2] = { "coherent_slave", "Coherent Slave" }, [SMCA_PIE] = { "pie", "Power, Interrupts, etc." }, [SMCA_UMC] = { "umc", "Unified Memory Controller" }, + [SMCA_UMC_V2] = { "umc_v2", "Unified Memory Controller" }, [SMCA_PB] = { "param_block", "Parameter Block" }, [SMCA_PSP] = { "psp", "Platform Security Processor" }, [SMCA_PSP_V2] = { "psp", "Platform Security Processor" },
@@ -98,6 +99,10 @@ static struct smca_bank_name smca_names[] = { [SMCA_MP5] = { "mp5", "Microprocessor 5 Unit" }, [SMCA_NBIO] = { "nbio", "Northbridge IO Unit" }, [SMCA_PCIE] = { "pcie", "PCI Express Unit" }, + [SMCA_PCIE_V2] = { "pcie", "PCI Express Unit" }, + [SMCA_XGMI_PCS] = { "xgmi_pcs", "Ext Global Memory Interconnect PCS Unit" }, + [SMCA_XGMI_PHY] = { "xgmi_phy", "Ext Global Memory Interconnect PHY Unit" }, + [SMCA_WAFL_PHY] = { "wafl_phy", "WAFL PHY Unit" }, };static const char *smca_get_name (enum smca_bank_types t)
@@ -155,6 +160,7 @@ static struct smca_hwid smca_hwid_mcatypes[] = {/* Unified Memory Controller MCA type */ { SMCA_UMC, HWID_MCATYPE(0x96, 0x0) }, + { SMCA_UMC_V2, HWID_MCATYPE(0x96, 0x1) },/* Parameter Block MCA type */ { SMCA_PB, HWID_MCATYPE(0x05, 0x0) },
@@ -175,6 +181,16 @@ static struct smca_hwid smca_hwid_mcatypes[] = {/* PCI Express Unit MCA type */ { SMCA_PCIE, HWID_MCATYPE(0x46, 0x0) }, + { SMCA_PCIE_V2, HWID_MCATYPE(0x46, 0x1) }, + +/* xGMI PCS MCA type */ + { SMCA_XGMI_PCS, HWID_MCATYPE(0x50, 0x0) }, + +/* xGMI PHY MCA type */ + { SMCA_XGMI_PHY, HWID_MCATYPE(0x259, 0x0) }, + +/* WAFL PHY MCA type */ + { SMCA_WAFL_PHY, HWID_MCATYPE(0x267, 0x0) }, };struct smca_bank smca_banks[MAX_NR_BANKS ];
drivers/edac/mce_amd.c +70 -0
94 lines not on this page · raw message
Signature · 1 line
2.17.1
PATCH 2/3 x86/MCE/AMD: Helper function to check UMC v2
From: Mukul Joshi <[email protected]>
Add a helper function to check if a given bank is
UMCv2 or not.
--- arch/x86/include/asm/mce.h | 2 ++ arch/x86/kernel/cpu/mce/amd.c | 6 ++++++ 2 files changed, 8 insertions(+)
arch/x86/include/asm/mce.h +2 -0
@@ -357,6 +357,7 @@ extern int mce_threshold_remove_device(unsigned int cpu);void mce_amd_feature_init (struct cpuinfo_x86 *c);int umc_normaddr_to_sysaddr (u64 norm_addr,u16 nid,u8 umc,u64 *sys_addr); +bool is_smca_umc_v2 (int bank);#else
@@ -366,6 +367,7 @@ static inline bool amd_mce_is_memory_error(struct mce *m) { return false; };static inline void mce_amd_feature_init (struct cpuinfo_x86 *c) { }static inline int umc_normaddr_to_sysaddr (u64 norm_addr,u16 nid,u8 umc,u64 *sys_addr) {return -EINVAL ; }; +static inline bool is_smca_umc_v2 (int bank) {return false; };#endif static inline void mce_hygon_feature_init (struct cpuinfo_x86 *c) {return mce_amd_feature_init (c); }
arch/x86/kernel/cpu/mce/amd.c +6 -0
@@ -1542,3 +1542,9 @@ int mce_threshold_create_device(unsigned int cpu)mce_threshold_remove_device (cpu);return err; } + +bool is_smca_umc_v2 (int bank) +{ +return (smca_get_bank_type (bank) ==SMCA_UMC_V2 ); +} +EXPORT_SYMBOL_GPL (is_smca_umc_v2);
Signature · 1 line
2.17.1
Borislav Petkov
maintainerRe: [PATCH 2/3] x86/MCE/AMD: Helper function to check UMC v2
Quoted 1 line · Naveen Krishna Chatradhi
Subject: Re: [PATCH 2/3] x86/MCE/AMD: Helper function to check UMC v2
The condensed patch description in the subject line should start with a uppercase letter and should be written in imperative tone:
"x86/MCE/AMD: Add a helper function... "
Quoted 3 lines
Signed-off-by Mukul Joshi <[email protected]>Reviewed-by John Clements <[email protected]>Reviewed-by Yazen Ghannam <[email protected]>
This is all fine and dandy but it needs your SOB too when you send the patch.
Please read Documentation/process/submitting-patches.rst
sections
Sign your work - the Developer's Certificate of Origin When to use Acked-by:, Cc:, and Co-developed-by:
arch/x86/kernel/cpu/mce/amd.c int mce_threshold_create_device(unsigned int cpu) · line 1550
Quoted 5 lines
arch/x86/kernel/cpu/mce/amd.c +5
-0+bool is_smca_umc_v2 (int bank) +{ +return (smca_get_bank_type (bank) ==SMCA_UMC_V2 ); +} +EXPORT_SYMBOL_GPL (is_smca_umc_v2);
This addition looks useless when it doesn't have any users.
Also, I'm pretty sceptical this even makes sense to have it exported - I'm guessing this is for mce_amd.c but I can't say without seeing it in use.
The same remarks hold true for your patch 3.
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteRE: [PATCH 2/3] x86/MCE/AMD: Helper function to check UMC v2
[AMD Official Use Only - Internal Distribution Only]
Hi Boris,
Thanks for the review comments.
I will update the Subject line and send out an updated patch.
In the meantime, here is the link to where the changes in Patch 2 and 3 are being used:
https://lists.freedesktop.org/archives/amd-gfx/2021-May/063423.html
Thanks.
Regards,
Mukul
-----Original Message-----
From: Borislav Petkov <[email protected]>
Sent: Tuesday, May 11, 2021 1:35 PM
To: Chatradhi, Naveen Krishna <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; Joshi, Mukul <[email protected]>
Subject: Re: [PATCH 2/3] x86/MCE/AMD: Helper function to check UMC v2
[CAUTION: External Email]
Quoted 1 line · Naveen Krishna Chatradhi
Subject: Re: [PATCH 2/3] x86/MCE/AMD: Helper function to check UMC v2
The condensed patch description in the subject line should start with a uppercase letter and should be written in imperative tone:
"x86/MCE/AMD: Add a helper function... "
Quoted 3 lines
Signed-off-by Mukul Joshi <[email protected]>Reviewed-by John Clements <[email protected]>Reviewed-by Yazen Ghannam <[email protected]>
This is all fine and dandy but it needs your SOB too when you send the patch.
Please read Documentation/process/submitting-patches.rst
sections
Sign your work - the Developer's Certificate of Origin When to use Acked-by:, Cc:, and Co-developed-by:
arch/x86/kernel/cpu/mce/amd.c int mce_threshold_create_device(unsigned int cpu) · line 1550
Quoted 4 lines
arch/x86/kernel/cpu/mce/amd.c +4
-0+bool is_smca_umc_v2 (int bank) +{ +return (smca_get_bank_type (bank) ==SMCA_UMC_V2 ); } +EXPORT_SYMBOL_GPL (is_smca_umc_v2);
This addition looks useless when it doesn't have any users.
Also, I'm pretty sceptical this even makes sense to have it exported - I'm guessing this is for mce_amd.c but I can't say without seeing it in use.
The same remarks hold true for your patch 3.
Thx.
--
Regards/Gruss,
Boris.
Borislav Petkov
maintainerRe: [PATCH 2/3] x86/MCE/AMD: Helper function to check UMC v2
Hi Mukul,
Quoted 1 line · Mukul
[AMD Official Use Only - Internal Distribution Only]
first of all, please do not top-post.
Quoted 2 lines
Thanks for the review comments.
I will update the Subject line and send out an updated patch.
Ok.
Quoted 2 lines
In the meantime, here is the link to where the changes in Patch 2 and 3 are being used:
https://lists.freedesktop.org/archives/amd-gfx/2021-May/063423.html
Yes, in the future, please Cc [email protected] too when you're using/exporting facilities to be used by other subsystems.
I'll try to reply to that patch through lore.kernel.org now.
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquettePATCH v2 1/2 x86/MCE/AMD: Export smca_get_bank_type symbol
In reply to Naveen Krishna ChatradhiExport smca_get_bank_type for use in the AMD GPU driver to determine MCA bank while handling correctable and uncorrectable errors in GPU UMC.
v1->v2:
- Drop the function is_smca_umc_v2().
- Drop the patch to introduce a new MCE priority (MCE_PRIO_ACEL) for GPU/accelarator cards.
--- arch/x86/include/asm/mce.h | 2 +- arch/x86/kernel/cpu/mce/amd.c | 3 ++- 2 files changed, 3 insertions(+), 2 deletions(-)
arch/x86/include/asm/mce.h +1 -1
@@ -358,7 +358,7 @@ extern int mce_threshold_remove_device(unsigned int cpu);void mce_amd_feature_init (struct cpuinfo_x86 *c);int umc_normaddr_to_sysaddr (u64 norm_addr,u16 nid,u8 umc,u64 *sys_addr); - +enum smca_bank_types smca_get_bank_type (unsigned int bank);#else static inline int mce_threshold_create_device (unsigned int cpu) {return 0 ; };
arch/x86/kernel/cpu/mce/amd.c +2 -1
@@ -123,7 +123,7 @@ const char *smca_get_long_name(enum smca_bank_types t) }EXPORT_SYMBOL_GPL (smca_get_long_name); -static enum smca_bank_types smca_get_bank_type (unsigned int bank) +enum smca_bank_types smca_get_bank_type (unsigned int bank) {struct smca_bank *b;
@@ -136,6 +136,7 @@ static enum smca_bank_types smca_get_bank_type(unsigned int bank)return b->hwid->bank_type; } +EXPORT_SYMBOL_GPL (smca_get_bank_type);static struct smca_hwid smca_hwid_mcatypes[] = {/* { bank_type, hwid_mcatype } */
Signature · 1 line
2.17.1
PATCH v2 2/2 drm/amdgpu: Register MCE notifier for Aldebaran RAS
On Aldebaran, GPU driver will handle bad page retirement even though UMC is host managed. As a result, register a bad page retirement handler on the mce notifier chain to retire bad pages on Aldebaran.
v1->v2:
- Use smca_get_bank_type() to determine MCA bank.
- Envelope the changes under #ifdef CONFIG_X86_MCE_AMD.
- Use MCE_PRIORITY_UC instead of MCE_PRIO_ACCEL as we are only handling uncorrectable errors.
- Use macros to determine UMC instance and channel instance where the uncorrectable error occured.
- Update the headline.
--- drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 142 ++++++++++++++++++++++++ 1 file changed, 142 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +142 -0
@@ -35,7 +35,11 @@#include "amdgpu_xgmi.h" #include "ivsrcid/nbio/irqsrcs_nbif_7_4.h" #include "atom.h" +#ifdef CONFIG_X86_MCE_AMD +#include <asm/mce.h> +static bool notifier_registered; +#endif static const char *RAS_FS_NAME ="ras" ;const char *ras_error_string[] = {
@@ -86,6 +90,9 @@ static bool amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con, uint64_t addr);static bool amdgpu_ras_check_bad_page (struct amdgpu_device *adev,uint64_t addr); +#ifdef CONFIG_X86_MCE_AMD +static void amdgpu_register_bad_pages_mca_notifier (void ); +#endif void amdgpu_ras_set_error_query_ready (struct amdgpu_device *adev,bool ready) {
@@ -2018,6 +2025,11 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) adev->smu.ppt_funcs->send_hbm_bad_pages_num (&adev->smu, con->eeprom_control.ras_num_recs); } +#ifdef CONFIG_X86_MCE_AMD +if ((adev->asic_type ==CHIP_ALDEBARAN ) && + (adev->gmc.xgmi.connected_to_cpu)) +amdgpu_register_bad_pages_mca_notifier (); +#endif return 0 ; free:
@@ -2511,3 +2523,133 @@ void amdgpu_release_ras_context(struct amdgpu_device *adev)kfree (con); } } + +#ifdef CONFIG_X86_MCE_AMD +static struct amdgpu_device *find_adev (uint32_t node_id) +{ +struct amdgpu_gpu_instance *gpu_instance; +int i; +struct amdgpu_device *adev =NULL ; + +mutex_lock (&mgpu_info.mutex); + +for (i =0 ; i < mgpu_info.num_gpu; i++) { + gpu_instance = &(mgpu_info.gpu_ins[i]); + adev = gpu_instance->adev; + +if (adev->gmc.xgmi.connected_to_cpu && + adev->gmc.xgmi.physical_node_id == node_id) +break ; + adev =NULL ; + } + +mutex_unlock (&mgpu_info.mutex); + +return adev; +} + +#define GET_MCA_IPID_GPUID (m) (((m) >> 44) & 0xF) +#define GET_UMC_INST (m) (((m) >> 21) & 0x7) +#define GET_CHAN_INDEX (m) ((((m) >> 12) & 0x3) | (((m) >> 18) & 0x4)) +#define GPU_ID_OFFSET 8 + +static int amdgpu_bad_page_notifier (struct notifier_block *nb, +unsigned long val,void *data) +{ +struct mce *m = (struct mce *)data; +struct amdgpu_device *adev =NULL ; +uint32_t gpu_id =0 ; +uint32_t umc_inst =0 ; +uint32_t ch_inst, channel_index =0 ; +struct ras_err_data err_data = {0 ,0 ,0 ,NULL }; +struct eeprom_table_record err_rec; +uint64_t retired_page; + +/* +* If the error was generated in UMC_V2, which belongs to GPU UMCs, +* and error occurred in DramECC (Extended error code = 0) then only +* process the error, else bail out. +*/ +if (!m || !((smca_get_bank_type (m->bank) ==SMCA_UMC_V2 ) && + (XEC (m->status,0x1f ) ==0x0 ))) +return NOTIFY_DONE ; + +/* +* GPU Id is offset by GPU_ID_OFFSET in MCA_IPID_UMC register. +*/ + gpu_id =GET_MCA_IPID_GPUID (m->ipid) -GPU_ID_OFFSET ; + + adev =find_adev (gpu_id); +if (!adev) { +dev_warn (adev->dev,"%s: Unable to find adev for gpu_id: %d\n" , + __func__, gpu_id); +return NOTIFY_DONE ; + } + +/* +* If it is correctable error, return. +*/ +if (mce_is_correctable (m)) { +return NOTIFY_OK ; + } + +/* +* If it is uncorrectable error, then find out UMC instance and +* channel index. +*/ + umc_inst =GET_UMC_INST (m->ipid); + ch_inst =GET_CHAN_INDEX (m->ipid); + +dev_info (adev->dev,"Uncorrectable error detected in UMC inst: %d, chan_idx: %d" , + umc_inst, ch_inst); + +memset (&err_rec,0x0 ,sizeof (struct eeprom_table_record )); + +/* +* Translate UMC channel address to Physical address +*/ + channel_index = + adev->umc.channel_idx_tbl[umc_inst * adev->umc.channel_inst_num + + ch_inst]; + + retired_page =ADDR_OF_8KB_BLOCK (m->addr) | +ADDR_OF_256B_BLOCK (channel_index) | +OFFSET_IN_256B_BLOCK (m->addr); + + err_rec.address = m->addr; + err_rec.retired_page = retired_page >>AMDGPU_GPU_PAGE_SHIFT ; + err_rec.ts = (uint64_t )ktime_get_real_seconds (); + err_rec.err_type =AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE ; + err_rec.cu =0 ; + err_rec.mem_channel = channel_index; + err_rec.mcumc_id = umc_inst; + + err_data.err_addr = &err_rec; + err_data.err_addr_cnt =1 ; + +if (amdgpu_bad_page_threshold !=0 ) { +amdgpu_ras_add_bad_pages (adev, err_data.err_addr, + err_data.err_addr_cnt); +amdgpu_ras_save_bad_pages (adev); + } + +return NOTIFY_OK ; +} + +static struct notifier_block amdgpu_bad_page_nb = { + .notifier_call = amdgpu_bad_page_notifier, + .priority =MCE_PRIO_UC , +}; + +static void amdgpu_register_bad_pages_mca_notifier (void ) +{ +/* +* Register the x86 notifier only once +* with MCE subsystem. +*/ +if (notifier_registered == false) { +mce_register_decode_chain (&amdgpu_bad_page_nb); + notifier_registered = true; + } +} +#endif
Signature · 1 line
2.17.1
Borislav Petkov
maintainerRe: [PATCHv2 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS
Quoted 5 lines · Mukul Joshi
On Aldebaran, GPU driver will handle bad page retirement even though UMC is host managed. As a result, register a bad page retirement handler on the mce notifier chain to retire bad pages on Aldebaran.
Quoted 8 lines
v1->v2:
- Usesmca_get_bank_type()to determine MCA bank.
- Envelope the changes under #ifdefCONFIG_X86_MCE_AMD.
- UseMCE_PRIORITY_UCinstead ofMCE_PRIO_ACCELas we are only handling uncorrectable errors.
- Use macros to determine UMC instance and channel instance where the uncorrectable error occured.
- Update the headline.
Same note as for the previous patch.
Quoted 29 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +29
-0+static int amdgpu_bad_page_notifier (struct notifier_block *nb, +unsigned long val,void *data) +{ +struct mce *m = (struct mce *)data; +struct amdgpu_device *adev =NULL ; +uint32_t gpu_id =0 ; +uint32_t umc_inst =0 ; +uint32_t ch_inst, channel_index =0 ; +struct ras_err_data err_data = {0 ,0 ,0 ,NULL }; +struct eeprom_table_record err_rec; +uint64_t retired_page; + +/* +* If the error was generated in UMC_V2, which belongs to GPU UMCs, +* and error occurred in DramECC (Extended error code = 0) then only +* process the error, else bail out. +*/ +if (!m || !((smca_get_bank_type (m->bank) ==SMCA_UMC_V2 ) && + (XEC (m->status,0x1f ) ==0x0 ))) +return NOTIFY_DONE ; + +/* +* GPU Id is offset by GPU_ID_OFFSET in MCA_IPID_UMC register. +*/ + gpu_id =GET_MCA_IPID_GPUID (m->ipid) -GPU_ID_OFFSET ; + + adev =find_adev (gpu_id); +if (!adev) { + dev_warn(adev->dev, "%s: Unable to find adev for gpu_id: %d\n",
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c void amdgpu_release_ras_context(struct amdgpu_device *adev) · line 2594
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +10
-0+ __func__, gpu_id); +return NOTIFY_DONE ; + } + +/* +* If it is correctable error, return. +*/ +if (mce_is_correctable (m)) { +return NOTIFY_OK ; + }
This can run before you find_adev().
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c void amdgpu_release_ras_context(struct amdgpu_device *adev) · line 2653
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +10
-0+static void amdgpu_register_bad_pages_mca_notifier (void ) +{ +/* +* Register the x86 notifier only once +* with MCE subsystem. +*/ +if (notifier_registered == false) { +mce_register_decode_chain (&amdgpu_bad_page_nb); + notifier_registered = true; + }
I have a patchset which will get rid of the need to do this silliness - only if I had some time to actually prepare it for submission... :-\
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteRE: [PATCHv2 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS
[AMD Official Use Only]
Quoted 26 lines
-----Original Message-----
From: Borislav Petkov <[email protected]>
Sent: Wednesday, September 22, 2021 7:41 AM
To: Joshi, Mukul <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; Ghannam, Yazen <[email protected]>; [email protected]
Subject: Re: [PATCHv2 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS[CAUTION: External Email]
Quoted 4 lines · Mukul Joshi
On Aldebaran, GPU driver will handle bad page retirement even though UMC is host managed. As a result, register a bad page retirement handler on the mce notifier chain to retire bad pages on Aldebaran.
Quoted 8 lines
v1->v2:
- Usesmca_get_bank_type()to determine MCA bank.
- Envelope the changes under #ifdefCONFIG_X86_MCE_AMD.
- UseMCE_PRIORITY_UCinstead ofMCE_PRIO_ACCELas we are only handling uncorrectable errors.
- Use macros to determine UMC instance and channel instance where the uncorrectable error occured.
- Update the headline.
Quoted 2 lines
Same note as for the previous patch.
Acked.
Quoted 32 lines
Quoted 32 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +32
-0+static int amdgpu_bad_page_notifier (struct notifier_block *nb, +unsigned long val,void *data) { +struct mce *m = (struct mce *)data; +struct amdgpu_device *adev =NULL ; +uint32_t gpu_id =0 ; +uint32_t umc_inst =0 ; +uint32_t ch_inst, channel_index =0 ; +struct ras_err_data err_data = {0 ,0 ,0 ,NULL }; +struct eeprom_table_record err_rec; +uint64_t retired_page; + +/* +* If the error was generated in UMC_V2, which belongs to GPU UMCs, +* and error occurred in DramECC (Extended error code = 0) then only +* process the error, else bail out. +*/ +if (!m || !((smca_get_bank_type (m->bank) ==SMCA_UMC_V2 ) && + (XEC (m->status,0x1f ) ==0x0 ))) +return NOTIFY_DONE ; + +/* +* GPU Id is offset by GPU_ID_OFFSET in MCA_IPID_UMC register. +*/ + gpu_id =GET_MCA_IPID_GPUID (m->ipid) -GPU_ID_OFFSET ; + + adev =find_adev (gpu_id); +if (!adev) { +dev_warn (adev->dev,"%s: Unable to find adev for gpu_id: %d\n" , + __func__, gpu_id); +return NOTIFY_DONE ; + } +
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c void amdgpu_release_ras_context(struct amdgpu_device *adev) · line 2594
Quoted 6 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +6
-0+/* +* If it is correctable error, return. +*/ +if (mce_is_correctable (m)) { +return NOTIFY_OK ; + }This can run before you
find_adev().
Acked.
Quoted 5 lines
Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +5
-0+static void amdgpu_register_bad_pages_mca_notifier (void ) +{ +/* +* Register the x86 notifier only once +* with MCE subsystem.
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c void amdgpu_release_ras_context(struct amdgpu_device *adev) · line 2653
Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +5
-0+*/ +if (notifier_registered == false) { +mce_register_decode_chain (&amdgpu_bad_page_nb); + notifier_registered = true; + }I have a patchset which will get rid of the need to do this silliness - only if I had some time to actually prepare it for submission... :-\
:-\
Thank you.
Regards,
Mukul
Quoted 11 lines
--
Regards/Gruss,
Boris.https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpeople.
kernel.org%2Ftglx%2Fnotes-about- netiquette&data=04%7C01%7Cmukul.joshi%40amd.com%7C7ae9c87153f7 4572712908d97dbdcc7a%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0 %7C637679076423976867%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAw MDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata =M5UDrSea0lnEi5%2BhU4ck0zk1dZD9kX4DUoXt95J6dJ4%3D&reserved=0
PATCH v3 2/2 drm/amdgpu: Register MCE notifier for Aldebaran RAS
In reply to Mukul JoshiOn Aldebaran, GPU driver will handle bad page retirement even though UMC is host managed. As a result, register a bad page retirement handler on the mce notifier chain to retire bad pages on Aldebaran.
---v1->v2:
- Use smca_get_bank_type() to determine MCA bank.
- Envelope the changes under #ifdef CONFIG_X86_MCE_AMD.
- Use MCE_PRIORITY_UC instead of MCE_PRIO_ACCEL as we are only handling uncorrectable errors.
- Use macros to determine UMC instance and channel instance where the uncorrectable error occured.
v2->v3:
- Move the check for correctable error before find_adev().
- Fix a NULL pointer dereference if find_adev() returns NULL.
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 141 ++++++++++++++++++++++++ 1 file changed, 141 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +141 -0
@@ -35,7 +35,11 @@#include "amdgpu_xgmi.h" #include "ivsrcid/nbio/irqsrcs_nbif_7_4.h" #include "atom.h" +#ifdef CONFIG_X86_MCE_AMD +#include <asm/mce.h> +static bool notifier_registered; +#endif static const char *RAS_FS_NAME ="ras" ;const char *ras_error_string[] = {
@@ -107,6 +111,9 @@ static bool amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con, uint64_t addr);static bool amdgpu_ras_check_bad_page (struct amdgpu_device *adev,uint64_t addr); +#ifdef CONFIG_X86_MCE_AMD +static void amdgpu_register_bad_pages_mca_notifier (void ); +#endif void amdgpu_ras_set_error_query_ready (struct amdgpu_device *adev,bool ready) {
@@ -2089,6 +2096,11 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) adev->smu.ppt_funcs->send_hbm_bad_pages_num (&adev->smu, con->eeprom_control.ras_num_recs); } +#ifdef CONFIG_X86_MCE_AMD +if ((adev->asic_type ==CHIP_ALDEBARAN ) && + (adev->gmc.xgmi.connected_to_cpu)) +amdgpu_register_bad_pages_mca_notifier (); +#endif return 0 ; free:
@@ -2583,3 +2595,132 @@ void amdgpu_release_ras_context(struct amdgpu_device *adev)kfree (con); } } + +#ifdef CONFIG_X86_MCE_AMD +static struct amdgpu_device *find_adev (uint32_t node_id) +{ +struct amdgpu_gpu_instance *gpu_instance; +int i; +struct amdgpu_device *adev =NULL ; + +mutex_lock (&mgpu_info.mutex); + +for (i =0 ; i < mgpu_info.num_gpu; i++) { + gpu_instance = &(mgpu_info.gpu_ins[i]); + adev = gpu_instance->adev; + +if (adev->gmc.xgmi.connected_to_cpu && + adev->gmc.xgmi.physical_node_id == node_id) +break ; + adev =NULL ; + } + +mutex_unlock (&mgpu_info.mutex); + +return adev; +} + +#define GET_MCA_IPID_GPUID (m) (((m) >> 44) & 0xF) +#define GET_UMC_INST (m) (((m) >> 21) & 0x7) +#define GET_CHAN_INDEX (m) ((((m) >> 12) & 0x3) | (((m) >> 18) & 0x4)) +#define GPU_ID_OFFSET 8 + +static int amdgpu_bad_page_notifier (struct notifier_block *nb, +unsigned long val,void *data) +{ +struct mce *m = (struct mce *)data; +struct amdgpu_device *adev =NULL ; +uint32_t gpu_id =0 ; +uint32_t umc_inst =0 ; +uint32_t ch_inst, channel_index =0 ; +struct ras_err_data err_data = {0 ,0 ,0 ,NULL }; +struct eeprom_table_record err_rec; +uint64_t retired_page; + +/* +* If the error was generated in UMC_V2, which belongs to GPU UMCs, +* and error occurred in DramECC (Extended error code = 0) then only +* process the error, else bail out. +*/ +if (!m || !((smca_get_bank_type (m->bank) ==SMCA_UMC_V2 ) && + (XEC (m->status,0x1f ) ==0x0 ))) +return NOTIFY_DONE ; + +/* +* If it is correctable error, return. +*/ +if (mce_is_correctable (m)) +return NOTIFY_OK ; + +/* +* GPU Id is offset by GPU_ID_OFFSET in MCA_IPID_UMC register. +*/ + gpu_id =GET_MCA_IPID_GPUID (m->ipid) -GPU_ID_OFFSET ; + + adev =find_adev (gpu_id); +if (!adev) { +DRM_WARN ("%s: Unable to find adev for gpu_id: %d\n" , __func__, + gpu_id); +return NOTIFY_DONE ; + } + +/* +* If it is uncorrectable error, then find out UMC instance and +* channel index. +*/ + umc_inst =GET_UMC_INST (m->ipid); + ch_inst =GET_CHAN_INDEX (m->ipid); + +dev_info (adev->dev,"Uncorrectable error detected in UMC inst: %d, chan_idx: %d" , + umc_inst, ch_inst); + +memset (&err_rec,0x0 ,sizeof (struct eeprom_table_record )); + +/* +* Translate UMC channel address to Physical address +*/ + channel_index = + adev->umc.channel_idx_tbl[umc_inst * adev->umc.channel_inst_num + + ch_inst]; + + retired_page =ADDR_OF_8KB_BLOCK (m->addr) | +ADDR_OF_256B_BLOCK (channel_index) | +OFFSET_IN_256B_BLOCK (m->addr); + + err_rec.address = m->addr; + err_rec.retired_page = retired_page >>AMDGPU_GPU_PAGE_SHIFT ; + err_rec.ts = (uint64_t )ktime_get_real_seconds (); + err_rec.err_type =AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE ; + err_rec.cu =0 ; + err_rec.mem_channel = channel_index; + err_rec.mcumc_id = umc_inst; + + err_data.err_addr = &err_rec; + err_data.err_addr_cnt =1 ; + +if (amdgpu_bad_page_threshold !=0 ) { +amdgpu_ras_add_bad_pages (adev, err_data.err_addr, + err_data.err_addr_cnt); +amdgpu_ras_save_bad_pages (adev); + } + +return NOTIFY_OK ; +} + +static struct notifier_block amdgpu_bad_page_nb = { + .notifier_call = amdgpu_bad_page_notifier, + .priority =MCE_PRIO_UC , +}; + +static void amdgpu_register_bad_pages_mca_notifier (void ) +{ +/* +* Register the x86 notifier only once +* with MCE subsystem. +*/ +if (notifier_registered == false) { +mce_register_decode_chain (&amdgpu_bad_page_nb); + notifier_registered = true; + } +} +#endif
Signature · 1 line
2.17.1
Yazen Ghannam
maintainerRe: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS
Quoted 5 lines · Mukul Joshi
On Aldebaran, GPU driver will handle bad page retirement even though UMC is host managed. As a result, register a bad page retirement handler on the mce notifier chain to retire bad pages on Aldebaran.
I think this should state that the driver will do page retirement for GPU-managed memory. As written, it implies that the driver do page retirement in general for the system.
...
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +10
-0+ +static int amdgpu_bad_page_notifier (struct notifier_block *nb, +unsigned long val,void *data) +{ +struct mce *m = (struct mce *)data; +struct amdgpu_device *adev =NULL ; +uint32_t gpu_id =0 ; +uint32_t umc_inst =0 ; +uint32_t ch_inst, channel_index =0 ; +struct ras_err_data err_data = {0 ,0 ,0 ,NULL };
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c void amdgpu_release_ras_context(struct amdgpu_device *adev) · line 2646
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +10
-0+struct eeprom_table_record err_rec; +uint64_t retired_page; + +/* +* If the error was generated in UMC_V2, which belongs to GPU UMCs, +* and error occurred in DramECC (Extended error code = 0) then only +* process the error, else bail out. +*/ +if (!m || !((smca_get_bank_type (m->bank) ==SMCA_UMC_V2 ) && + (XEC (m->status,0x1f ) ==0x0 )))
The MCA_STATUS[ErrorCodeExt] field is bits [21:16], so the mask should be 0x3f.
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c void amdgpu_release_ras_context(struct amdgpu_device *adev) · line 2653
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +7
-0+return NOTIFY_DONE ; + +/* +* If it is correctable error, return. +*/ +if (mce_is_correctable (m)) +return NOTIFY_OK ;
Shouldn't this be "NOTIFY_DONE" if "don't care" about this error?
Thanks,
Yazen
Borislav Petkov
maintainerRe: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS
Quoted 11 lines · Yazen Ghannam
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c void amdgpu_release_ras_context(struct amdgpu_device *adev) · line 2646
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +7
-0+/* +* If the error was generated in UMC_V2, which belongs to GPU UMCs, +* and error occurred in DramECC (Extended error code = 0) then only +* process the error, else bail out. +*/ +if (!m || !((smca_get_bank_type (m->bank) ==SMCA_UMC_V2 ) && + (XEC (m->status,0x1f ) ==0x0 )))The MCA_STATUS[ErrorCodeExt] field is bits [21:16], so the mask should be
0x3f.
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c void amdgpu_release_ras_context(struct amdgpu_device *adev) · line 2653
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +7
-0+return NOTIFY_DONE ; + +/* +* If it is correctable error, return. +*/ +if (mce_is_correctable (m)) +return NOTIFY_OK ;Shouldn't this be "
NOTIFY_DONE" if "don't care" about this error?
I think the logic here is to stop calling any further consumers on the notify chain because this is a GPU correctable error and they can't do anything about it anyway, right?
Or am I misreading it?
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteRE: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS
[AMD Official Use Only]
Quoted 32 lines
-----Original Message-----
From: Borislav Petkov <[email protected]>
Sent: Thursday, September 23, 2021 10:37 AM
To: Ghannam, Yazen <[email protected]>
Cc: Joshi, Mukul <[email protected]>; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]
Subject: Re: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS[CAUTION: External Email]
Quoted 19 lines · Yazen Ghannam
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c void amdgpu_release_ras_context(struct amdgpu_device *adev) · line 2646
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +7
-0+/* +* If the error was generated in UMC_V2, which belongs to GPU UMCs, +* and error occurred in DramECC (Extended error code = 0) then only +* process the error, else bail out. +*/ +if (!m || !((smca_get_bank_type (m->bank) ==SMCA_UMC_V2 ) && + (XEC (m->status,0x1f ) ==0x0 )))The MCA_STATUS[ErrorCodeExt] field is bits [21:16], so the mask should be
0x3f.Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +7
-0+return NOTIFY_DONE ; + +/* +* If it is correctable error, return. +*/ +if (mce_is_correctable (m)) +return NOTIFY_OK ;
Quoted 5 lines
Quoted 1 line
Shouldn't this be "
NOTIFY_DONE" if "don't care" about this error?I think the logic here is to stop calling any further consumers on the notify chain because this is a GPU correctable error and they can't do anything about it anyway, right?
That's correct.
Regards,
Mukul
Quoted 14 lines
Or am I misreading it?
--
Regards/Gruss,
Boris.https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpeople.
kernel.org%2Ftglx%2Fnotes-about- netiquette&data=04%7C01%7Cmukul.joshi%40amd.com%7C1493bcdae62 8466d5b8408d97e9fa6f0%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0 %7C637680046455034969%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAw MDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata =uqbjJslUPJabZonJc6m9Ub3C5IkZDPdqwr6MI0oLPcc%3D&reserved=0
RE: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS
In reply to Yazen Ghannam[AMD Official Use Only]
Quoted 11 lines
-----Original Message-----
From: Ghannam, Yazen <[email protected]>
Sent: Thursday, September 23, 2021 10:29 AM
To: Joshi, Mukul <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; amd- [email protected]
Subject: Re: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RASOn Wed, Sep 22, 2021 at 03:36:20PM -0400, Mukul Joshi wrote:
Quoted 9 lines
Quoted 4 lines
On Aldebaran, GPU driver will handle bad page retirement even though UMC is host managed. As a result, register a bad page retirement handler on the mce notifier chain to retire bad pages on Aldebaran.
I think this should state that the driver will do page retirement for GPU-managed memory. As written, it implies that the driver do page retirement in general for the system.
ACK. I will update the description.
Quoted 16 lines
...
Quoted 14 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +14
-0+ +static int amdgpu_bad_page_notifier (struct notifier_block *nb, +unsigned long val,void *data) { +struct mce *m = (struct mce *)data; +struct amdgpu_device *adev =NULL ; +uint32_t gpu_id =0 ; +uint32_t umc_inst =0 ; +uint32_t ch_inst, channel_index =0 ; +struct ras_err_data err_data = {0 ,0 ,0 ,NULL }; +struct eeprom_table_record err_rec; +uint64_t retired_page; + +/* +* If the error was generated in UMC_V2, which belongs to GPU UMCs,
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c void amdgpu_release_ras_context(struct amdgpu_device *adev) · line 2646
Quoted 5 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +5
-0+* and error occurred in DramECC (Extended error code = 0) then only +* process the error, else bail out. +*/ +if (!m || !((smca_get_bank_type (m->bank) ==SMCA_UMC_V2 ) && + (XEC (m->status,0x1f ) ==0x0 )))The MCA_STATUS[ErrorCodeExt] field is bits [21:16], so the mask should be
0x3f.
Ack. Thanks for catching this.
Quoted 10 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c void amdgpu_release_ras_context(struct amdgpu_device *adev) · line 2653
Quoted 8 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +7
-0+return NOTIFY_DONE ; + +/* +* If it is correctable error, return. +*/ +if (mce_is_correctable (m)) +return NOTIFY_OK ;Shouldn't this be "
NOTIFY_DONE" if "don't care" about this error?
The thinking is we want to stop calling further consumers since it's a correctable error in GPU UMC and we are not taking any action about the correctable errors.
Thanks,
Mukul
Quoted 3 lines
Thanks,
Yazen
Yazen Ghannam
maintainerRe: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS
On Thu, Sep 23, 2021 at 11:30:55AM -0400, Joshi, Mukul wrote:
...
Quoted 11 lines
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c void amdgpu_release_ras_context(struct amdgpu_device *adev) · line 2653
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +7
-0+return NOTIFY_DONE ; + +/* +* If it is correctable error, return. +*/ +if (mce_is_correctable (m)) +return NOTIFY_OK ;Shouldn't this be "
NOTIFY_DONE" if "don't care" about this error?The thinking is we want to stop calling further consumers since it's a correctable error in GPU UMC and we are not taking any action about the correctable errors.
Shouldn't the error still be reported to EDAC for decoding and counting? I think users want this.
But it looks to me that either NOTIFY_OK or NOTIFY_DONE will allow this, so it's not a big deal. Was this intended to be NOTIFY_STOP?
Thanks,
Yazen
Borislav Petkov
maintainerRe: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS
Quoted 2 lines · Yazen Ghannam
Shouldn't the error still be reported to EDAC for decoding and counting? I think users want this.
You know what happens with users getting ECCs reported, right? They think immediately their hw is going bad and start wanting to replace it...
So what does actually tell you if you were a simple user and you had 5 correctable errors in the GPU VRAM?
All you wanna do is play, I'd say.
:-)
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteYazen Ghannam
maintainerRe: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS
Quoted 11 lines · Borislav Petkov
Quoted 2 lines · Yazen Ghannam
Shouldn't the error still be reported to EDAC for decoding and counting? I think users want this.
You know what happens with users getting ECCs reported, right? They think immediately their hw is going bad and start wanting to replace it...
So what does actually tell you if you were a simple user and you had 5 correctable errors in the GPU VRAM?
I agree with you in general. But this device isn't really a GPU. And users of this device seem to want to count *every* error, at least for now.
Quoted 4 lines
All you wanna do is play, I'd say.
:-)
Definitely. :)
Thanks,
Yazen
Borislav Petkov
maintainerRe: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS
Quoted 3 lines · Yazen Ghannam
I agree with you in general. But this device isn't really a GPU. And users of this device seem to want to count *every* error, at least for now.
Aha, so something accelerator-y where they do general purpose computation.
So what's the big picture here: they count all the errors and when they reach a certain amount, they decide to replace the GPUs just in case?
Or wait until they become uncorrectable? But then it doesn't matter because we will handle it properly by excluding the VRAM range from further use.
Or do they wanna see *when* they had the correctable errors so that they can restart the computation, just in case.
Dunno, it would be a lot helpful if we had some RAS strategy for those things...
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteYazen Ghannam
maintainerRe: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS
Quoted 14 lines · Borislav Petkov
Quoted 3 lines · Yazen Ghannam
I agree with you in general. But this device isn't really a GPU. And users of this device seem to want to count *every* error, at least for now.
Aha, so something accelerator-y where they do general purpose computation.
So what's the big picture here: they count all the errors and when they reach a certain amount, they decide to replace the GPUs just in case?
Or wait until they become uncorrectable? But then it doesn't matter because we will handle it properly by excluding the VRAM range from further use.
Quoted 6 lines
Or do they wanna see *when* they had the correctable errors so that they can restart the computation, just in case.
Dunno, it would be a lot helpful if we had some RAS strategy for those things...
I completely agree. The system integrators have their own policies for error tracking, part replacement, etc. I expect they'll propose kernel changes if they want any. Though I think general strategies will become apparent once these sort of devices are in wider use.
Thanks,
Yazen
RE: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS
In reply to Yazen Ghannam[AMD Official Use Only]
Quoted 22 lines
-----Original Message-----
From: Ghannam, Yazen <[email protected]>
Sent: Thursday, September 23, 2021 1:23 PM
To: Joshi, Mukul <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; amd- [email protected]
Subject: Re: [PATCHv3 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RASOn Thu, Sep 23, 2021 at 11:30:55AM -0400, Joshi, Mukul wrote:
...Quoted 10 lines
Quoted 9 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c void amdgpu_release_ras_context(struct amdgpu_device *adev) · line 2653
Quoted 7 lines
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +7
-0+return NOTIFY_DONE ; + +/* +* If it is correctable error, return. +*/ +if (mce_is_correctable (m)) +return NOTIFY_OK ;Shouldn't this be "
NOTIFY_DONE" if "don't care" about this error?
Quoted 4 lines
Quoted 1 line
The thinking is we want to stop calling further consumers since it's a
correctable error in GPU UMC and we are not taking any action about the correctable errors.
Sorry I have to retract this back a bit. I remembered I started with the intention Of using NOTIFY_STOP but realized that we would not be doing any accounting in this function.
Quoted 6 lines
Shouldn't the error still be reported to EDAC for decoding and counting? I think users want this.
But it looks to me that either
NOTIFY_OKorNOTIFY_DONEwill allow this, so it's not a big deal. Was this intended to beNOTIFY_STOP?
Sorry I have to retract my previous comment about stopping further consumers a bit.
I remembered I started with the intention to use NOTIFY_STOP but realized we were not doing any accounting in this function. Later I guess I went by the comments put against NOTIFY_OK in notifier.h:
#define NOTIFY_DONE 0x0000 /* Don't care */
#define NOTIFY_OK 0x0001 /* Suits me */
Because this was a correctable error on GPU UMC, NOTIFY_OK ("Suits me") was probably more suited to this condition, even though we were not taking any action on the correctable errors.
Thanks,
Mukul
Quoted 2 lines
Thanks,
Yazen
PATCH v4 2/2 drm/amdgpu: Register MCE notifier for Aldebaran RAS
In reply to Mukul JoshiOn Aldebaran, GPU driver will handle bad page retirement for GPU memory even though UMC is host managed. As a result, register a bad page retirement handler on the mce notifier chain to retire bad pages on Aldebaran.
---v1->v2:
- Use smca_get_bank_type() to determine MCA bank.
- Envelope the changes under #ifdef CONFIG_X86_MCE_AMD.
- Use MCE_PRIORITY_UC instead of MCE_PRIO_ACCEL as we are only handling uncorrectable errors.
- Use macros to determine UMC instance and channel instance where the uncorrectable error occured.
v2->v3:
- Move the check for correctable error before find_adev().
- Fix a NULL pointer dereference if find_adev() returns NULL.
v3->v4:
- Update the commit log to specify page retirement for GPU memory only.
- Fix the mask passed to XEC macro.
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c | 141 ++++++++++++++++++++++++ 1 file changed, 141 insertions(+)
drivers/gpu/drm/amd/amdgpu/amdgpu_ras.c +141 -0
@@ -35,7 +35,11 @@#include "amdgpu_xgmi.h" #include "ivsrcid/nbio/irqsrcs_nbif_7_4.h" #include "atom.h" +#ifdef CONFIG_X86_MCE_AMD +#include <asm/mce.h> +static bool notifier_registered; +#endif static const char *RAS_FS_NAME ="ras" ;const char *ras_error_string[] = {
@@ -107,6 +111,9 @@ static bool amdgpu_ras_check_bad_page_unlock(struct amdgpu_ras *con, uint64_t addr);static bool amdgpu_ras_check_bad_page (struct amdgpu_device *adev,uint64_t addr); +#ifdef CONFIG_X86_MCE_AMD +static void amdgpu_register_bad_pages_mca_notifier (void ); +#endif void amdgpu_ras_set_error_query_ready (struct amdgpu_device *adev,bool ready) {
@@ -2089,6 +2096,11 @@ int amdgpu_ras_recovery_init(struct amdgpu_device *adev) adev->smu.ppt_funcs->send_hbm_bad_pages_num (&adev->smu, con->eeprom_control.ras_num_recs); } +#ifdef CONFIG_X86_MCE_AMD +if ((adev->asic_type ==CHIP_ALDEBARAN ) && + (adev->gmc.xgmi.connected_to_cpu)) +amdgpu_register_bad_pages_mca_notifier (); +#endif return 0 ; free:
@@ -2552,3 +2564,132 @@ void amdgpu_release_ras_context(struct amdgpu_device *adev)kfree (con); } } + +#ifdef CONFIG_X86_MCE_AMD +static struct amdgpu_device *find_adev (uint32_t node_id) +{ +struct amdgpu_gpu_instance *gpu_instance; +int i; +struct amdgpu_device *adev =NULL ; + +mutex_lock (&mgpu_info.mutex); + +for (i =0 ; i < mgpu_info.num_gpu; i++) { + gpu_instance = &(mgpu_info.gpu_ins[i]); + adev = gpu_instance->adev; + +if (adev->gmc.xgmi.connected_to_cpu && + adev->gmc.xgmi.physical_node_id == node_id) +break ; + adev =NULL ; + } + +mutex_unlock (&mgpu_info.mutex); + +return adev; +} + +#define GET_MCA_IPID_GPUID (m) (((m) >> 44) & 0xF) +#define GET_UMC_INST (m) (((m) >> 21) & 0x7) +#define GET_CHAN_INDEX (m) ((((m) >> 12) & 0x3) | (((m) >> 18) & 0x4)) +#define GPU_ID_OFFSET 8 + +static int amdgpu_bad_page_notifier (struct notifier_block *nb, +unsigned long val,void *data) +{ +struct mce *m = (struct mce *)data; +struct amdgpu_device *adev =NULL ; +uint32_t gpu_id =0 ; +uint32_t umc_inst =0 ; +uint32_t ch_inst, channel_index =0 ; +struct ras_err_data err_data = {0 ,0 ,0 ,NULL }; +struct eeprom_table_record err_rec; +uint64_t retired_page; + +/* +* If the error was generated in UMC_V2, which belongs to GPU UMCs, +* and error occurred in DramECC (Extended error code = 0) then only +* process the error, else bail out. +*/ +if (!m || !((smca_get_bank_type (m->bank) ==SMCA_UMC_V2 ) && + (XEC (m->status,0x3f ) ==0x0 ))) +return NOTIFY_DONE ; + +/* +* If it is correctable error, return. +*/ +if (mce_is_correctable (m)) +return NOTIFY_OK ; + +/* +* GPU Id is offset by GPU_ID_OFFSET in MCA_IPID_UMC register. +*/ + gpu_id =GET_MCA_IPID_GPUID (m->ipid) -GPU_ID_OFFSET ; + + adev =find_adev (gpu_id); +if (!adev) { +DRM_WARN ("%s: Unable to find adev for gpu_id: %d\n" , __func__, + gpu_id); +return NOTIFY_DONE ; + } + +/* +* If it is uncorrectable error, then find out UMC instance and +* channel index. +*/ + umc_inst =GET_UMC_INST (m->ipid); + ch_inst =GET_CHAN_INDEX (m->ipid); + +dev_info (adev->dev,"Uncorrectable error detected in UMC inst: %d, chan_idx: %d" , + umc_inst, ch_inst); + +memset (&err_rec,0x0 ,sizeof (struct eeprom_table_record )); + +/* +* Translate UMC channel address to Physical address +*/ + channel_index = + adev->umc.channel_idx_tbl[umc_inst * adev->umc.channel_inst_num + + ch_inst]; + + retired_page =ADDR_OF_8KB_BLOCK (m->addr) | +ADDR_OF_256B_BLOCK (channel_index) | +OFFSET_IN_256B_BLOCK (m->addr); + + err_rec.address = m->addr; + err_rec.retired_page = retired_page >>AMDGPU_GPU_PAGE_SHIFT ; + err_rec.ts = (uint64_t )ktime_get_real_seconds (); + err_rec.err_type =AMDGPU_RAS_EEPROM_ERR_NON_RECOVERABLE ; + err_rec.cu =0 ; + err_rec.mem_channel = channel_index; + err_rec.mcumc_id = umc_inst; + + err_data.err_addr = &err_rec; + err_data.err_addr_cnt =1 ; + +if (amdgpu_bad_page_threshold !=0 ) { +amdgpu_ras_add_bad_pages (adev, err_data.err_addr, + err_data.err_addr_cnt); +amdgpu_ras_save_bad_pages (adev); + } + +return NOTIFY_OK ; +} + +static struct notifier_block amdgpu_bad_page_nb = { + .notifier_call = amdgpu_bad_page_notifier, + .priority =MCE_PRIO_UC , +}; + +static void amdgpu_register_bad_pages_mca_notifier (void ) +{ +/* +* Register the x86 notifier only once +* with MCE subsystem. +*/ +if (notifier_registered == false) { +mce_register_decode_chain (&amdgpu_bad_page_nb); + notifier_registered = true; + } +} +#endif
Signature · 1 line
2.17.1
Re: [PATCHv4 2/2] drm/amdgpu: Register MCE notifier for Aldebaran RAS
Quoted 7 lines · Mukul Joshi
On Aldebaran, GPU driver will handle bad page retirement for GPU memory even though UMC is host managed. As a result, register a bad page retirement handler on the mce notifier chain to retire bad pages on Aldebaran.
Signed-off-by Mukul Joshi <[email protected]>---
This patch looks good to me overall.
Thanks,
Yazen
Borislav Petkov
maintainerRe: [PATCHv2 1/2] x86/MCE/AMD: Export smca_get_bank_type symbol
In reply to Mukul JoshiQuoted 8 lines · Mukul Joshi
Export
smca_get_bank_typefor use in the AMD GPU driver to determine MCA bank while handling correctable and uncorrectable errors in GPU UMC.
v1->v2:
- Drop the functionis_smca_umc_v2().
- Drop the patch to introduce a new MCE priority (MCE_PRIO_ACEL) for GPU/accelarator cards.
Patch changelog information goes...
Quoted 3 lines
Signed-off-by Mukul Joshi <[email protected]>---
... under this line so that it gets automatically removed by git when applying the patch.
Alex, how do you wanna handle this?
Want me to ACK this and you can carry it through your tree along with the second patch?
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteDeucher, Alexander
maintainerRE: [PATCHv2 1/2] x86/MCE/AMD: Export smca_get_bank_type symbol
[Public]
Quoted 29 lines
-----Original Message-----
From: amd-gfx <[email protected]> On Behalf Of Borislav Petkov
Sent: Wednesday, September 22, 2021 7:34 AM
To: Joshi, Mukul <[email protected]>; Alex Deucher <[email protected]>
Cc: [email protected]; [email protected]; linux- [email protected]; [email protected]; [email protected]; Ghannam, Yazen <[email protected]>; amd- [email protected]
Subject: Re: [PATCHv2 1/2] x86/MCE/AMD: Exportsmca_get_bank_typesymbolQuoted 8 lines · Mukul Joshi
Export
smca_get_bank_typefor use in the AMD GPU driver to determine MCA bank while handling correctable and uncorrectable errors in GPU UMC.
v1->v2:
- Drop the functionis_smca_umc_v2().
- Drop the patch to introduce a new MCE priority (MCE_PRIO_ACEL) for GPU/accelarator cards.Patch changelog information goes...
Quoted 3 lines
Signed-off-by Mukul Joshi <[email protected]>---
Quoted 7 lines
... under this line so that it gets automatically removed by git when applying the patch.
Alex, how do you wanna handle this?
Want me to ACK this and you can carry it through your tree along with the second patch?
That would be great. Thanks!
Alex
Quoted 13 lines
--
Regards/Gruss,
Boris.https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpeo
ple.kernel.org%2Ftglx%2Fnotes-about- netiquette&data=04%7C01%7Calexander.deucher%40amd.com%7C12b cf4eeffad4e2533b508d97dca1cf4%7C3dd8961fe4884e608e11a82d994e183d% 7C0%7C0%7C637679129761221057%7CUnknown%7CTWFpbGZsb3d8eyJWIjoi MC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C200 0&sdata=tK9aK%2FHf5RimF%2FenuTGeJSFFmRuk86Q%2BqY9Jt23gKMQ %3D&reserved=0
Re: [PATCHv2 1/2] x86/MCE/AMD: Export smca_get_bank_type symbol
Quoted 19 lines · Alexander
Quoted 19 lines · Mukul Joshi
Quoted 8 lines
Export
smca_get_bank_typefor use in the AMD GPU driver to determine MCA bank while handling correctable and uncorrectable errors in GPU UMC.
v1->v2:
- Drop the functionis_smca_umc_v2().
- Drop the patch to introduce a new MCE priority (MCE_PRIO_ACEL) for GPU/accelarator cards.Patch changelog information goes...
Quoted 3 lines
Signed-off-by Mukul Joshi <[email protected]>---
... under this line so that it gets automatically removed by git when applying the patch.
Quoted 6 lines
Quoted 4 lines
Alex, how do you wanna handle this?
Want me to ACK this and you can carry it through your tree along with the second patch?
That would be great. Thanks!
Ok, with the above changelog removed:
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteRE: [PATCHv2 1/2] x86/MCE/AMD: Export smca_get_bank_type symbol
[AMD Official Use Only]
Quoted 37 lines
-----Original Message-----
From: Borislav Petkov <[email protected]>
Sent: Wednesday, September 22, 2021 12:44 PM
To: Deucher, Alexander <[email protected]>
Cc: Joshi, Mukul <[email protected]>; Alex Deucher <[email protected]>; [email protected]; [email protected]; linux- [email protected]; [email protected]; [email protected]; Ghannam, Yazen <[email protected]>; [email protected]
Subject: Re: [PATCHv2 1/2] x86/MCE/AMD: Exportsmca_get_bank_typesymbol[CAUTION: External Email]
Quoted 24 lines · Alexander
Quoted 23 lines · Mukul Joshi
Quoted 8 lines
Export
smca_get_bank_typefor use in the AMD GPU driver to determine MCA bank while handling correctable and uncorrectable errors in GPU UMC.
v1->v2:
- Drop the functionis_smca_umc_v2().
- Drop the patch to introduce a new MCE priority (MCE_PRIO_ACEL) for GPU/accelarator cards.Patch changelog information goes...
Quoted 3 lines
Signed-off-by Mukul Joshi <[email protected]>---
... under this line so that it gets automatically removed by git when applying the patch.
Alex, how do you wanna handle this?
Want me to ACK this and you can carry it through your tree along with the second patch?
Quoted 8 lines
Quoted 1 line
That would be great. Thanks!
Ok, with the above changelog removed:
Acked-by Borislav Petkov <[email protected]>Thx.
Thank you so much!
I will make sure to remove the changelog.
And I will send the updated version for the second patch soon.
Regards,
Mukul
Quoted 12 lines
--
Regards/Gruss,
Boris.https://nam11.safelinks.protection.outlook.com/?url=https%3A%2F%2Fpeople.
kernel.org%2Ftglx%2Fnotes-about- netiquette&data=04%7C01%7CMukul.Joshi%40amd.com%7C3dc61ec5018f 487ec06e08d97de83039%7C3dd8961fe4884e608e11a82d994e183d%7C0%7C0 %7C637679258473597532%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAw MDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata =4JqFDJpxM%2Bzl4%2BZoaC3tTwScEhRy2Aa7xJaNJn3rxbE%3D&reserved= 0
PATCH 3/3 x86/mce: Add MCE priority for Accelerator devices
In reply to Naveen Krishna ChatradhiFrom: Mukul Joshi <[email protected]>
Create a new MCE priority for accelerator devices on the notifier chain. On some future AMD platforms, GPU devices will process MCE errors and work as error detection devices.
--- arch/x86/include/asm/mce.h | 1 + 1 file changed, 1 insertion(+)
arch/x86/include/asm/mce.h +1 -0
@@ -173,6 +173,7 @@ enum mce_notifier_prios {MCE_PRIO_LOWEST ,MCE_PRIO_MCELOG ,MCE_PRIO_EDAC , +MCE_PRIO_ACCEL ,MCE_PRIO_NFIT ,MCE_PRIO_EXTLOG , MCE_PRIO_UC,
Signature · 1 line
2.17.1
Borislav Petkov
maintainerarch/x86/kernel/cpu/mce/amd.c static struct smca_bank_name smca_names[] = { · line 93
Quoted 9 lines · Naveen Krishna Chatradhi
arch/x86/kernel/cpu/mce/amd.c +1
-0@@ -90,6 +90,7 @@ static struct smca_bank_name smca_names[] = { [SMCA_CS_V2] = { "coherent_slave", "Coherent Slave" }, [SMCA_PIE] = { "pie", "Power, Interrupts, etc." }, [SMCA_UMC] = { "umc", "Unified Memory Controller" }, + [SMCA_UMC_V2] = { "umc_v2", "Unified Memory Controller" },
So this is called "umc_v2" but the other V2 FUs's strings are the same. Why?
Also, if you're going to repeat strings, you can just as well group all those which are the same this way:
[ SMCA_UMC ... SMCA_UMC_V2 ] = { "umc", "Unified Memory Controller" },
and do that for all which have V1 and V2.
I mean, gcc is smart enough to do that behind the scenes for identical strings but you should do that in C too.
drivers/edac/mce_amd.c static const char * const smca_umc_mce_desc[] = { · line 326
Quoted 9 lines
drivers/edac/mce_amd.c +1
-0@@ -323,6 +323,21 @@ static const char * const smca_umc_mce_desc[] = { "AES SRAM ECC error", }; +static const char *const smca_umc2_mce_desc[] = {
Ok, gcc reuses the identical string pointers from smca_umc_mce_desc[] so we should be ok wrt duplication.
Quoted 13 lines
drivers/edac/mce_amd.c +13
-0+"DRAM ECC error" , +"Data poison error" , +"SDP parity error" , +"Reserved" , +"Address/Command parity error" , +"Write data parity error" , +"DCQ SRAM ECC error" , +"Reserved" , +"Read data parity error" , +"Rdb SRAM ECC error" , +"RdRsp SRAM ECC error" , + "LM32 MP errors", +};
...
Quoted 20 lines
drivers/edac/mce_amd.c +20
-0+static const char *const smca_xgmipcs_mce_desc[] = { +"DataLossErr" , +"TrainingErr" , +"FlowCtrlAckErr" , +"RxFifoUnderflowErr" , +"RxFifoOverflowErr" , +"CRCErr" , +"BERExceededErr" , +"TxVcidDataErr" , +"ReplayBufParityErr" , +"DataParityErr" , +"ReplayFifoOverflowErr" , +"ReplayFIfoUnderflowErr" , +"ElasticFifoOverflowErr" , +"DeskewErr" , +"FlowCtrlCRCErr" , +"DataStartupLimitErr" , +"FCInitTimeoutErr" , +"RecoveryTimeoutErr" , +"ReadySerialTimeoutErr" ,
drivers/edac/mce_amd.c static const char * const smca_pcie_mce_desc[] = { · line 451
Quoted 10 lines
drivers/edac/mce_amd.c +10
-0+"ReadySerialAttemptErr" , +"RecoveryAttemptErr" , +"RecoveryRelockAttemptErr" , +"ReplayAttemptErr" , +"SyncHdrErr" , +"TxReplayTimeoutErr" , +"RxReplayTimeoutErr" , +"LinkSubTxTimeoutErr" , +"LinkSubRxTimeoutErr" , + "RxCMDPktErr",
What happened to those and why aren't they proper words like the other error descriptions?
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquette[AMD Official Use Only]
Hi Boris
My apologies for delayed response. Thanks for your review comments, will submit a v2 shortly.
Regards,
Naveenk
-----Original Message-----
From: Borislav Petkov <[email protected]>
Sent: Tuesday, May 11, 2021 10:57 PM
To: Chatradhi, Naveen Krishna <[email protected]>
Cc: [email protected]; [email protected]; [email protected]; [email protected]; [email protected]; M K, Muralidhara <[email protected]>
Subject: Re: [PATCH 1/3] x86/MCE/AMD, EDAC/mce_amd: Add new SMCA bank types.
[CAUTION: External Email]
arch/x86/kernel/cpu/mce/amd.c static struct smca_bank_name smca_names[] = { · line 93
Quoted 10 lines · Naveen Krishna Chatradhi
(quoted diff) +0
-0
b/arch/x86/kernel/cpu/mce/amd.cindexe486f96b3cb3..055f3a0acf5e100644arch/x86/kernel/cpu/mce/amd.c +1
-0@@ -90,6 +90,7 @@ static struct smca_bank_name smca_names[] = { [SMCA_CS_V2] = { "coherent_slave", "Coherent Slave" }, [SMCA_PIE] = { "pie", "Power, Interrupts, etc." }, [SMCA_UMC] = { "umc", "Unified Memory Controller" }, + [SMCA_UMC_V2] = { "umc_v2", "Unified Memory Controller" },
So this is called "umc_v2" but the other V2 FUs's strings are the same.
Why?
[naveenk:] There is a possibility for a heterogenous system with both the SMCA_UMC and SMCA_UMC_V2 variant of controllers to exist. I will update the long name to describe accordingly.
Also, if you're going to repeat strings, you can just as well group all those which are the same this way:
[ SMCA_UMC ... SMCA_UMC_V2 ] = { "umc", "Unified Memory Controller" },
and do that for all which have V1 and V2.
I mean, gcc is smart enough to do that behind the scenes for identical strings but you should do that in C too. [naveenk:] thanks for the suggestion, I can do this for the other units.
drivers/edac/mce_amd.c static const char * const smca_umc_mce_desc[] = { · line 326
Quoted 9 lines
drivers/edac/mce_amd.c index renamed from drivers/edac/mce_amd.c +0
-0
5dd905a3f30c..5515fd9336b1100644drivers/edac/mce_amd.c +1
-0@@ -323,6 +323,21 @@ static const char * const smca_umc_mce_desc[] = { "AES SRAM ECC error", }; +static const char *const smca_umc2_mce_desc[] = {
Ok, gcc reuses the identical string pointers from smca_umc_mce_desc[] so we should be ok wrt duplication.
Quoted 13 lines
drivers/edac/mce_amd.c +13
-0+"DRAM ECC error" , +"Data poison error" , +"SDP parity error" , +"Reserved" , +"Address/Command parity error" , +"Write data parity error" , +"DCQ SRAM ECC error" , +"Reserved" , +"Read data parity error" , +"Rdb SRAM ECC error" , +"RdRsp SRAM ECC error" , + "LM32 MP errors", +};
...
Quoted 20 lines
drivers/edac/mce_amd.c +20
-0+static const char *const smca_xgmipcs_mce_desc[] = { +"DataLossErr" , +"TrainingErr" , +"FlowCtrlAckErr" , +"RxFifoUnderflowErr" , +"RxFifoOverflowErr" , +"CRCErr" , +"BERExceededErr" , +"TxVcidDataErr" , +"ReplayBufParityErr" , +"DataParityErr" , +"ReplayFifoOverflowErr" , +"ReplayFIfoUnderflowErr" , +"ElasticFifoOverflowErr" , +"DeskewErr" , +"FlowCtrlCRCErr" , +"DataStartupLimitErr" , +"FCInitTimeoutErr" , +"RecoveryTimeoutErr" , +"ReadySerialTimeoutErr" ,
drivers/edac/mce_amd.c static const char * const smca_pcie_mce_desc[] = { · line 451
Quoted 10 lines
drivers/edac/mce_amd.c +10
-0+"ReadySerialAttemptErr" , +"RecoveryAttemptErr" , +"RecoveryRelockAttemptErr" , +"ReplayAttemptErr" , +"SyncHdrErr" , +"TxReplayTimeoutErr" , +"RxReplayTimeoutErr" , +"LinkSubTxTimeoutErr" , +"LinkSubRxTimeoutErr" , + "RxCMDPktErr",
What happened to those and why aren't they proper words like the other error descriptions? [naveenk:] Will change these into proper words.
Thx.
--
Regards/Gruss,
Boris.
Borislav Petkov
maintainerQuoted 2 lines · Naveen Krishna
[naveenk:] There is a possibility for a heterogenous system with both the
SMCA_UMCandSMCA_UMC_V2variant of controllers to exist.
Wait, what? You can have systems with *both* UMCs in the same coherent fabric and thus the OS can see UMCs of both types on the same system?
Thx.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquetteYazen Ghannam
maintainerQuoted 7 lines · Borislav Petkov
Quoted 2 lines · Naveen Krishna
[naveenk:] There is a possibility for a heterogenous system with both the
SMCA_UMCandSMCA_UMC_V2variant of controllers to exist.Wait, what? You can have systems with *both* UMCs in the same coherent fabric and thus the OS can see UMCs of both types on the same system?
Yep, that's right. The UMCs are the only case of this so far and in the foreseeable future. Though we may be moving towards more cases like this.
Thanks,
Yazen
Borislav Petkov
maintainerQuoted 3 lines · Yazen Ghannam
Yep, that's right. The UMCs are the only case of this so far and in the foreseeable future. Though we may be moving towards more cases like this.
Ok, then you guys really need to make sure those strings are unique because they're in sysfs:
/sys/devices/system/machinecheck/machinecheck...
AFAIR, so we can't have duplicates there.
Signature · 4 lines
Regards/Gruss,
Boris.
https://people.kernel.org/tglx/notes-about-netiquettePATCH v2 1/1 x86/MCE/AMD, EDAC/mce_amd: Add new SMCA bank types.
In reply to Naveen Krishna ChatradhiFrom: Muralidhara M K <[email protected]>
Add the (HWID, MCATYPE) tuples and names for new
SMCA bank types.
Also, add their respective error descriptions to the MCE decoding module edac_mce_amd.
Also while at it, optimize the string names for some SMCA banks.
--- arch/x86/include/asm/mce.h | 5 +++ arch/x86/kernel/cpu/mce/amd.c | 55 ++++++++++++++++----------- drivers/edac/mce_amd.c | 70 +++++++++++++++++++++++++++++++++++ 3 files changed, 109 insertions(+), 21 deletions(-)
arch/x86/include/asm/mce.h +5 -0
@@ -317,6 +317,7 @@ enum smca_bank_types {SMCA_CS_V2 ,/* Coherent Slave */ SMCA_PIE ,/* Power, Interrupts, etc. */ SMCA_UMC ,/* Unified Memory Controller */ +SMCA_UMC_V2 ,/* Unified Memory Controller V2 */ SMCA_PB ,/* Parameter Block */ SMCA_PSP ,/* Platform Security Processor */ SMCA_PSP_V2, /* Platform Security Processor */
@@ -325,6 +326,10 @@ enum smca_bank_types {SMCA_MP5 ,/* Microprocessor 5 Unit */ SMCA_NBIO ,/* Northbridge IO Unit */ SMCA_PCIE ,/* PCI Express Unit */ +SMCA_PCIE_V2 ,/* PCI Express Unit */ +SMCA_XGMI_PCS ,/* xGMI PCS Unit */ +SMCA_XGMI_PHY ,/* xGMI PHY Unit */ +SMCA_WAFL_PHY ,/* WAFL PHY Unit */ N_SMCA_BANK_TYPES };
arch/x86/kernel/cpu/mce/amd.c +34 -21
73 lines not on this page · raw message
drivers/edac/mce_amd.c +70 -0
94 lines not on this page · raw message
Signature · 1 line
2.26.2
[tip: ras/core] x86/MCE/AMD, EDAC/mce_amd: Add new SMCA bank types
The following commit has been merged into the ras/core branch of tip:
The following commit has been merged into the ras/core branch of tip:
Commit-ID: 94a311ce248e0b53c76e110fd00511af47b72ffb Gitweb: https://git.kernel.org/tip/94a311ce248e0b53c76e110fd00511af47b72ffb Author: Muralidhara M K <[email protected]> AuthorDate: Wed, 26 May 2021 22:16:01 +05:30 Committer: Borislav Petkov <[email protected]> CommitterDate: Thu, 27 May 2021 20:08:14 +02:00
x86/MCE/AMD, EDAC/mce_amd: Add new SMCA bank types
Add the (HWID, MCATYPE) tuples and names for new SMCA bank types.
Also, add their respective error descriptions to the MCE decoding module edac_mce_amd. Also while at it, optimize the string names for some SMCA banks.
[ bp: Drop repeated comments, explain why UMC_V2 is a separate entry. ]
--- arch/x86/include/asm/mce.h | 13 ++++-- arch/x86/kernel/cpu/mce/amd.c | 55 ++++++++++++++++----------- drivers/edac/mce_amd.c | 70 ++++++++++++++++++++++++++++++++++- 3 files changed, 113 insertions(+), 25 deletions(-)
arch/x86/include/asm/mce.h +9 -4
@@ -305,7 +305,7 @@ extern void apei_mce_report_mem_error(int corrected,/* These may be used by multiple smca_hwid_mcatypes */ enum smca_bank_types {SMCA_LS =0 ,/* Load Store */ -SMCA_LS_V2 ,/* Load Store */ +SMCA_LS_V2 ,SMCA_IF ,/* Instruction Fetch */ SMCA_L2_CACHE ,/* L2 Cache */ SMCA_DE ,/* Decoder Unit */
@@ -314,17 +314,22 @@ enum smca_bank_types {SMCA_FP ,/* Floating Point */ SMCA_L3_CACHE ,/* L3 Cache */ SMCA_CS ,/* Coherent Slave */ -SMCA_CS_V2 ,/* Coherent Slave */ +SMCA_CS_V2 ,SMCA_PIE ,/* Power, Interrupts, etc. */ SMCA_UMC ,/* Unified Memory Controller */ +SMCA_UMC_V2 ,SMCA_PB ,/* Parameter Block */ SMCA_PSP ,/* Platform Security Processor */ -SMCA_PSP_V2 ,/* Platform Security Processor */ +SMCA_PSP_V2 ,SMCA_SMU ,/* System Management Unit */ -SMCA_SMU_V2 ,/* System Management Unit */ +SMCA_SMU_V2 ,SMCA_MP5 ,/* Microprocessor 5 Unit */ SMCA_NBIO ,/* Northbridge IO Unit */ SMCA_PCIE ,/* PCI Express Unit */ +SMCA_PCIE_V2 , +SMCA_XGMI_PCS ,/* xGMI PCS Unit */ +SMCA_XGMI_PHY ,/* xGMI PHY Unit */ +SMCA_WAFL_PHY ,/* WAFL PHY Unit */ N_SMCA_BANK_TYPES };
arch/x86/kernel/cpu/mce/amd.c +34 -21
73 lines not on this page · raw message
drivers/edac/mce_amd.c +70 -0
94 lines not on this page · raw message