[PATCH 0/2] EDAC/amd64: Consolidate Family 1Ah setup and fix UMC csrow decode

[PATCH 0/2] EDAC/amd64: Consolidate Family 1Ah setup and fix UMC csrow decode

[Vishal Badole]

Two small amd64_edac cleanups/fixes for AMD Family 1Ah.

Patch 1 sets zn_regs_v2 once at the Family 1Ah level instead of in each
model case, so future models cannot miss the v2 register layout. No
functional change.

Patch 2 masks the UMC chip select from MCA_SYND with 0x3 instead of 0x7.
Only four chip selects are ever implemented, so the old mask could decode
an out-of-range csrow and trip an EDAC core warning.

Vishal Badole (2):
  EDAC/amd64: Set zn_regs_v2 for all Family 1Ah models
  EDAC/amd64: Mask UMC chip select to the four implemented selects

 drivers/edac/amd64_edac.c | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)

-- 
2.34.1

[PATCH 1/2] EDAC/amd64: Set zn_regs_v2 for all Family 1Ah models

[Vishal Badole]

The v2 register layout applies to every Family 1Ah model, but
zn_regs_v2 is set individually in each model case of the switch. All
currently supported models set it, so this is not a live bug; however a
model added later without its own case, or in a case that omits the
assignment, would read the UMC registers at the wrong offsets.

Set zn_regs_v2 once at the Family 1Ah level so it applies to all models
uniformly, and drop the per-model assignments.

No functional change intended.

Suggested-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Vishal Badole <Vishal.Badole@amd.com>
---
 drivers/edac/amd64_edac.c | 8 ++------
 1 file changed, 2 insertions(+), 6 deletions(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index 475235c402e8..cab696be4f07 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -3887,23 +3887,19 @@ static int per_family_init(struct amd64_pvt *pvt)
 		break;
 
 	case 0x1A:
+		pvt->flags.zn_regs_v2   = 1;
+
 		switch (pvt->model) {
 		case 0x00 ... 0x1f:
 			pvt->max_mcs            = 12;
-			pvt->flags.zn_regs_v2   = 1;
-			break;
-		case 0x40 ... 0x4f:
-			pvt->flags.zn_regs_v2   = 1;
 			break;
 		case 0x50 ... 0x57:
 		case 0xc0 ... 0xc7:
 			pvt->max_mcs            = 16;
-			pvt->flags.zn_regs_v2   = 1;
 			break;
 		case 0x90 ... 0x9f:
 		case 0xa0 ... 0xaf:
 			pvt->max_mcs            = 8;
-			pvt->flags.zn_regs_v2   = 1;
 			break;
 		}
 		break;
-- 
2.34.1

[PATCH 2/2] EDAC/amd64: Mask UMC chip select to the four implemented selects

[Vishal Badole]

umc_get_err_info() masks the Chip Select from MCA_SYND with 0x7, a
leftover from Zen 1 where the architecture reserved space for up to
eight chip selects. Only four have ever been implemented, and newer
products document this as MCA_SYND_UMC[1:0] = Chip Select.

Since only four csrows are allocated, a syndrome with bit 2 set decodes
to an out-of-range csrow and trips:

  EDAC MC0: INTERNAL ERROR: csrow value is out of range (5 >= 4)

Mask with 0x3 to match the four implemented chip selects.

Suggested-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Vishal Badole <Vishal.Badole@amd.com>
---
 drivers/edac/amd64_edac.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
index cab696be4f07..b69c3f51dd37 100644
--- a/drivers/edac/amd64_edac.c
+++ b/drivers/edac/amd64_edac.c
@@ -2802,7 +2802,7 @@ static inline void decode_bus_error(int node_id, struct mce *m)
 static void umc_get_err_info(struct mce *m, struct err_info *err)
 {
 	err->channel = (m->ipid & GENMASK(31, 0)) >> 20;
-	err->csrow = m->synd & 0x7;
+	err->csrow = m->synd & 0x3;
 }
 
 static void decode_umc_error(int node_id, struct mce *m)
-- 
2.34.1

Re: [PATCH 2/2] EDAC/amd64: Mask UMC chip select to the four implemented selects

[Borislav Petkov]

On Fri, Sep 25, 2026 at 10:56:39PM +0530, Vishal Badole wrote:
> umc_get_err_info() masks the Chip Select from MCA_SYND with 0x7, a
> leftover from Zen 1 where the architecture reserved space for up to
> eight chip selects. Only four have ever been implemented, and newer
> products document this as MCA_SYND_UMC[1:0] = Chip Select.
> 
> Since only four csrows are allocated, a syndrome with bit 2 set decodes
> to an out-of-range csrow and trips:
> 
>   EDAC MC0: INTERNAL ERROR: csrow value is out of range (5 >= 4)
> 
> Mask with 0x3 to match the four implemented chip selects.
> 
> Suggested-by: Yazen Ghannam <yazen.ghannam@amd.com>
> Signed-off-by: Vishal Badole <Vishal.Badole@amd.com>
> ---
>  drivers/edac/amd64_edac.c | 2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
> index cab696be4f07..b69c3f51dd37 100644
> --- a/drivers/edac/amd64_edac.c
> +++ b/drivers/edac/amd64_edac.c
> @@ -2802,7 +2802,7 @@ static inline void decode_bus_error(int node_id, struct mce *m)
>  static void umc_get_err_info(struct mce *m, struct err_info *err)
>  {
>  	err->channel = (m->ipid & GENMASK(31, 0)) >> 20;
> -	err->csrow = m->synd & 0x7;
> +	err->csrow = m->synd & 0x3;
>  }
>  
>  static void decode_umc_error(int node_id, struct mce *m)
> -- 

Is this ever going to change?

Do we need to track this in the per-model struct, i.e., what your patch 1 is
touching?

Is this a fix that needs to go to stable? If so, Fixes: tag and CC:stable.

If so, patches going to stable go first and need to be minimal for ease of
backporting.

Thx.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette

Re: [PATCH 1/2] EDAC/amd64: Set zn_regs_v2 for all Family 1Ah models

[Yazen Ghannam]

On Fri, Sep 25, 2026 at 10:56:38PM +0530, Vishal Badole wrote:
> The v2 register layout applies to every Family 1Ah model, but
> zn_regs_v2 is set individually in each model case of the switch. All
> currently supported models set it, so this is not a live bug; however a
> model added later without its own case, or in a case that omits the
> assignment, would read the UMC registers at the wrong offsets.

This parapragh should be simpler. Also, please avoid using long, run-on,
compound sentences.

> 
> Set zn_regs_v2 once at the Family 1Ah level so it applies to all models
> uniformly, and drop the per-model assignments.
> 
> No functional change intended.
> 
> Suggested-by: Yazen Ghannam <yazen.ghannam@amd.com>
> Signed-off-by: Vishal Badole <Vishal.Badole@amd.com>
> ---
>  drivers/edac/amd64_edac.c | 8 ++------
>  1 file changed, 2 insertions(+), 6 deletions(-)
> 
> diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
> index 475235c402e8..cab696be4f07 100644
> --- a/drivers/edac/amd64_edac.c
> +++ b/drivers/edac/amd64_edac.c
> @@ -3887,23 +3887,19 @@ static int per_family_init(struct amd64_pvt *pvt)
>  		break;
>  
>  	case 0x1A:
> +		pvt->flags.zn_regs_v2   = 1;
> +
>  		switch (pvt->model) {
>  		case 0x00 ... 0x1f:
>  			pvt->max_mcs            = 12;
> -			pvt->flags.zn_regs_v2   = 1;
> -			break;
> -		case 0x40 ... 0x4f:
> -			pvt->flags.zn_regs_v2   = 1;
>  			break;
>  		case 0x50 ... 0x57:
>  		case 0xc0 ... 0xc7:
>  			pvt->max_mcs            = 16;
> -			pvt->flags.zn_regs_v2   = 1;
>  			break;
>  		case 0x90 ... 0x9f:
>  		case 0xa0 ... 0xaf:
>  			pvt->max_mcs            = 8;
> -			pvt->flags.zn_regs_v2   = 1;
>  			break;
>  		}
>  		break;
> -- 

Otherwise, looks good to me.

Reviewed-by: Yazen Ghannam <yazen.ghannam@amd.com>

Thanks,
Yazen

Re: [PATCH 2/2] EDAC/amd64: Mask UMC chip select to the four implemented selects

[Yazen Ghannam]

On Fri, Sep 25, 2026 at 11:14:29AM -0700, Borislav Petkov wrote:
> On Fri, Sep 25, 2026 at 10:56:39PM +0530, Vishal Badole wrote:
> > umc_get_err_info() masks the Chip Select from MCA_SYND with 0x7, a
> > leftover from Zen 1 where the architecture reserved space for up to
> > eight chip selects. Only four have ever been implemented, and newer
> > products document this as MCA_SYND_UMC[1:0] = Chip Select.
> > 
> > Since only four csrows are allocated, a syndrome with bit 2 set decodes
> > to an out-of-range csrow and trips:
> > 
> >   EDAC MC0: INTERNAL ERROR: csrow value is out of range (5 >= 4)
> > 
> > Mask with 0x3 to match the four implemented chip selects.
> > 
> > Suggested-by: Yazen Ghannam <yazen.ghannam@amd.com>
> > Signed-off-by: Vishal Badole <Vishal.Badole@amd.com>
> > ---
> >  drivers/edac/amd64_edac.c | 2 +-
> >  1 file changed, 1 insertion(+), 1 deletion(-)
> > 
> > diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
> > index cab696be4f07..b69c3f51dd37 100644
> > --- a/drivers/edac/amd64_edac.c
> > +++ b/drivers/edac/amd64_edac.c
> > @@ -2802,7 +2802,7 @@ static inline void decode_bus_error(int node_id, struct mce *m)
> >  static void umc_get_err_info(struct mce *m, struct err_info *err)
> >  {
> >  	err->channel = (m->ipid & GENMASK(31, 0)) >> 20;
> > -	err->csrow = m->synd & 0x7;
> > +	err->csrow = m->synd & 0x3;
> >  }
> >  
> >  static void decode_umc_error(int node_id, struct mce *m)
> > -- 
> 
> Is this ever going to change?
> 
> Do we need to track this in the per-model struct, i.e., what your patch 1 is
> touching?

This works for everything up through Family 1Ah.

It *could* change in the future, but I don't want to speculate.

Also, I'm working on a fresh EDAC module without all the model-specific
stuff. This is the sort of thing it would avoid.

> 
> Is this a fix that needs to go to stable? If so, Fixes: tag and CC:stable.
> 
> If so, patches going to stable go first and need to be minimal for ease of
> backporting.
> 

I didn't consider this a necessary fix. But it'd be good to have it to
avoid noise from reports.

Vishal, please do add the 'Fixes:' tag, Cc stable, and reorder the
patches.

Otherwise, looks good.

Reviewed-by: Yazen Ghannam <yazen.ghannam@amd.com>

Thanks,
Yazen

Re: [PATCH 1/2] EDAC/amd64: Set zn_regs_v2 for all Family 1Ah models

[Badole, Vishal]

On 9/28/2026 7:51 PM, Yazen Ghannam wrote:
> On Fri, Sep 25, 2026 at 10:56:38PM +0530, Vishal Badole wrote:
>> The v2 register layout applies to every Family 1Ah model, but
>> zn_regs_v2 is set individually in each model case of the switch. All
>> currently supported models set it, so this is not a live bug; however a
>> model added later without its own case, or in a case that omits the
>> assignment, would read the UMC registers at the wrong offsets.
> 
> This parapragh should be simpler. Also, please avoid using long, run-on,
> compound sentences.
> 
Sure, I will update it in next patch revision.>>
>> Set zn_regs_v2 once at the Family 1Ah level so it applies to all models
>> uniformly, and drop the per-model assignments.
>>
>> No functional change intended.
>>
>> Suggested-by: Yazen Ghannam <yazen.ghannam@amd.com>
>> Signed-off-by: Vishal Badole <Vishal.Badole@amd.com>
>> ---
>>   drivers/edac/amd64_edac.c | 8 ++------
>>   1 file changed, 2 insertions(+), 6 deletions(-)
>>
>> diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
>> index 475235c402e8..cab696be4f07 100644
>> --- a/drivers/edac/amd64_edac.c
>> +++ b/drivers/edac/amd64_edac.c
>> @@ -3887,23 +3887,19 @@ static int per_family_init(struct amd64_pvt *pvt)
>>   		break;
>>   
>>   	case 0x1A:
>> +		pvt->flags.zn_regs_v2   = 1;
>> +
>>   		switch (pvt->model) {
>>   		case 0x00 ... 0x1f:
>>   			pvt->max_mcs            = 12;
>> -			pvt->flags.zn_regs_v2   = 1;
>> -			break;
>> -		case 0x40 ... 0x4f:
>> -			pvt->flags.zn_regs_v2   = 1;
>>   			break;
>>   		case 0x50 ... 0x57:
>>   		case 0xc0 ... 0xc7:
>>   			pvt->max_mcs            = 16;
>> -			pvt->flags.zn_regs_v2   = 1;
>>   			break;
>>   		case 0x90 ... 0x9f:
>>   		case 0xa0 ... 0xaf:
>>   			pvt->max_mcs            = 8;
>> -			pvt->flags.zn_regs_v2   = 1;
>>   			break;
>>   		}
>>   		break;
>> -- 
> 
> Otherwise, looks good to me.
> 
> Reviewed-by: Yazen Ghannam <yazen.ghannam@amd.com>
> 
> Thanks,
> Yazen

Re: [PATCH 2/2] EDAC/amd64: Mask UMC chip select to the four implemented selects

[Badole, Vishal]

On 9/28/2026 8:08 PM, Yazen Ghannam wrote:
> On Fri, Sep 25, 2026 at 11:14:29AM -0700, Borislav Petkov wrote:
>> On Fri, Sep 25, 2026 at 10:56:39PM +0530, Vishal Badole wrote:
>>> umc_get_err_info() masks the Chip Select from MCA_SYND with 0x7, a
>>> leftover from Zen 1 where the architecture reserved space for up to
>>> eight chip selects. Only four have ever been implemented, and newer
>>> products document this as MCA_SYND_UMC[1:0] = Chip Select.
>>>
>>> Since only four csrows are allocated, a syndrome with bit 2 set decodes
>>> to an out-of-range csrow and trips:
>>>
>>>    EDAC MC0: INTERNAL ERROR: csrow value is out of range (5 >= 4)
>>>
>>> Mask with 0x3 to match the four implemented chip selects.
>>>
>>> Suggested-by: Yazen Ghannam <yazen.ghannam@amd.com>
>>> Signed-off-by: Vishal Badole <Vishal.Badole@amd.com>
>>> ---
>>>   drivers/edac/amd64_edac.c | 2 +-
>>>   1 file changed, 1 insertion(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/edac/amd64_edac.c b/drivers/edac/amd64_edac.c
>>> index cab696be4f07..b69c3f51dd37 100644
>>> --- a/drivers/edac/amd64_edac.c
>>> +++ b/drivers/edac/amd64_edac.c
>>> @@ -2802,7 +2802,7 @@ static inline void decode_bus_error(int node_id, struct mce *m)
>>>   static void umc_get_err_info(struct mce *m, struct err_info *err)
>>>   {
>>>   	err->channel = (m->ipid & GENMASK(31, 0)) >> 20;
>>> -	err->csrow = m->synd & 0x7;
>>> +	err->csrow = m->synd & 0x3;
>>>   }
>>>   
>>>   static void decode_umc_error(int node_id, struct mce *m)
>>> -- 
>>
>> Is this ever going to change?
>>
>> Do we need to track this in the per-model struct, i.e., what your patch 1 is
>> touching?
> 
> This works for everything up through Family 1Ah.
> 
> It *could* change in the future, but I don't want to speculate.
> 
> Also, I'm working on a fresh EDAC module without all the model-specific
> stuff. This is the sort of thing it would avoid.
> 
>>
>> Is this a fix that needs to go to stable? If so, Fixes: tag and CC:stable.
>>
>> If so, patches going to stable go first and need to be minimal for ease of
>> backporting.
>>
> 
> I didn't consider this a necessary fix. But it'd be good to have it to
> avoid noise from reports.
> 
> Vishal, please do add the 'Fixes:' tag, Cc stable, and reorder the
> patches.
> 
Sure, I will update it in next patch revision.

> Otherwise, looks good.
> 
> Reviewed-by: Yazen Ghannam <yazen.ghannam@amd.com>
> 
> Thanks,
> Yazen

Re: [PATCH 2/2] EDAC/amd64: Mask UMC chip select to the four implemented selects

[Borislav Petkov]

On Mon, Sep 28, 2026 at 10:38:21AM -0400, Yazen Ghannam wrote:
> I didn't consider this a necessary fix. But it'd be good to have it to
> avoid noise from reports.

I didn't either but what is Vishal doing to trigger this:

"Since only four csrows are allocated, a syndrome with bit 2 set decodes
to an out-of-range csrow and trips:

  EDAC MC0: INTERNAL ERROR: csrow value is out of range (5 >= 4)"

Vishal?

From reading the commit message, I still have no clue how that happens and
what are we fixing here.

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette

Re: [PATCH 2/2] EDAC/amd64: Mask UMC chip select to the four implemented selects

[Yazen Ghannam]

On Mon, Sep 28, 2026 at 12:20:45PM -0700, Borislav Petkov wrote:
> On Mon, Sep 28, 2026 at 10:38:21AM -0400, Yazen Ghannam wrote:
> > I didn't consider this a necessary fix. But it'd be good to have it to
> > avoid noise from reports.
> 
> I didn't either but what is Vishal doing to trigger this:
> 
> "Since only four csrows are allocated, a syndrome with bit 2 set decodes
> to an out-of-range csrow and trips:
> 
>   EDAC MC0: INTERNAL ERROR: csrow value is out of range (5 >= 4)"
> 
> Vishal?
> 
> From reading the commit message, I still have no clue how that happens and
> what are we fixing here.
> 

There are a couple of cases that trip this.

1) Newer hardware where 'bit 2' is redefined. For example, it could mean
'sub-channel' or something else relevant. In that case, a real error
would be decoded incorrectly.

2) Some hardware error injection methods that use arbitrary values. This
is like the 'error simulation' we do in the mce-inject module where we
can write something to the registers.

So the patch is a fix for #2 today, and it is a preemptive fix for #1.

Thanks,
Yazen