EDAC support for 'Rocket Lake'

EDAC support for 'Rocket Lake'

[Christian Lorenz]

Hi,

we are using the EDAC interface on Linux for our internal ECC DIMM
hardware monitoring.

With Sky Lake CPUs, the kernel driver in use is 'skx_edac'.

We have now received servers with the following CPUs, for which there
does not seem to be kernel EDAC support:

   processor       : 11
   vendor_id       : GenuineIntel
   cpu family      : 6
   model           : 167
   model name      : Intel(R) Xeon(R) E-2386G CPU @ 3.50GHz
   external link   : see [1] at the end of this mail

We have observed that there seems to be no EDAC support for this CPU,
/sys/devices/system/edac is not populated, tools like edac-utils or
rasdaemon do not function without it.

We have tested this with the same result on:

   - RHEL 8.6 -- kernel version 4.18.0-372
   - AlmaLinux 8.6 -- kernel version 4.18.0-372
   - RHEL 9.0 -- kernel version 5.14.0

According to documentation, the E-2386G codename is 'formerly Rocket Lake'.

Our question: Is there or will there be EDAC support for this 'Rocket Lake'
Xeon E-23XX CPU?

Thank you,
Christian


[1] 
https://ark.intel.com/content/www/us/en/ark/products/214806/intel-xeon-e2386g-processor-12m-cache-3-50-ghz.html

Re: EDAC support for 'Rocket Lake'

[Tony Luck]

On Wed, Oct 05, 2022 at 02:13:13PM +0200, Christian Lorenz wrote:
> Hi,
> 
> we are using the EDAC interface on Linux for our internal ECC DIMM
> hardware monitoring.
> 
> With Sky Lake CPUs, the kernel driver in use is 'skx_edac'.
> 
> We have now received servers with the following CPUs, for which there
> does not seem to be kernel EDAC support:
> 
>   processor       : 11
>   vendor_id       : GenuineIntel
>   cpu family      : 6
>   model           : 167
>   model name      : Intel(R) Xeon(R) E-2386G CPU @ 3.50GHz
>   external link   : see [1] at the end of this mail
> 
> We have observed that there seems to be no EDAC support for this CPU,
> /sys/devices/system/edac is not populated, tools like edac-utils or
> rasdaemon do not function without it.
> 
> We have tested this with the same result on:
> 
>   - RHEL 8.6 -- kernel version 4.18.0-372
>   - AlmaLinux 8.6 -- kernel version 4.18.0-372
>   - RHEL 9.0 -- kernel version 5.14.0
> 
> According to documentation, the E-2386G codename is 'formerly Rocket Lake'.
> 
> Our question: Is there or will there be EDAC support for this 'Rocket Lake'
> Xeon E-23XX CPU?

Christian,

Historically Intel has provided EDAC drivers for the "big" Xeon
systems (formery "-E5" and "-E7", currently "-SP" Xeon models).

Support for the Xeon-E3 models had been provided by members of
the community. Most recent update for the ie31200_edac driver
to support Kaby Lake was from Jason Baron (added to this e-mail).

Jason: Are you looking at Rocket Lake?

-Tony

Re: EDAC support for 'Rocket Lake'

[Jason Baron]

On 10/5/22 11:21, Tony Luck wrote:
> On Wed, Oct 05, 2022 at 02:13:13PM +0200, Christian Lorenz wrote:
>> Hi,
>>
>> we are using the EDAC interface on Linux for our internal ECC DIMM
>> hardware monitoring.
>>
>> With Sky Lake CPUs, the kernel driver in use is 'skx_edac'.
>>
>> We have now received servers with the following CPUs, for which there
>> does not seem to be kernel EDAC support:
>>
>>   processor       : 11
>>   vendor_id       : GenuineIntel
>>   cpu family      : 6
>>   model           : 167
>>   model name      : Intel(R) Xeon(R) E-2386G CPU @ 3.50GHz
>>   external link   : see [1] at the end of this mail
>>
>> We have observed that there seems to be no EDAC support for this CPU,
>> /sys/devices/system/edac is not populated, tools like edac-utils or
>> rasdaemon do not function without it.
>>
>> We have tested this with the same result on:
>>
>>   - RHEL 8.6 -- kernel version 4.18.0-372
>>   - AlmaLinux 8.6 -- kernel version 4.18.0-372
>>   - RHEL 9.0 -- kernel version 5.14.0
>>
>> According to documentation, the E-2386G codename is 'formerly Rocket Lake'.
>>
>> Our question: Is there or will there be EDAC support for this 'Rocket Lake'
>> Xeon E-23XX CPU?
> 
> Christian,
> 
> Historically Intel has provided EDAC drivers for the "big" Xeon
> systems (formery "-E5" and "-E7", currently "-SP" Xeon models).
> 
> Support for the Xeon-E3 models had been provided by members of
> the community. Most recent update for the ie31200_edac driver
> to support Kaby Lake was from Jason Baron (added to this e-mail).
> 
> Jason: Are you looking at Rocket Lake?
> 

Hi,

I've really only added hardware I have access to, and we don't have
'Rocket Lake'. I'm happy to help work on support if you have a datasheet
and can test. It may be as easy as adding some new pci ids for example:

7a14a11f93df EDAC/ie31200: Add Skylake-S support

Thanks,

-Jason

RE: EDAC support for 'Rocket Lake'

[Luck, Tony]

> I've really only added hardware I have access to, and we don't have
> 'Rocket Lake'. I'm happy to help work on support if you have a datasheet
> and can test. It may be as easy as adding some new pci ids for example:

Jason,

Some web searching hasn't turned up a public data sheet. The EDS volume 2
is Intel Doc # 615924. Do you have access through Akamai? If that confirms that
you just need to add some new PCI IDs etc. then I can get someone to make
it OK to post a patch.

Testing is harder. I can probably get access to a rocket lake internally. But
I don't think those support ACPI error injection. So would only be able to basic
"does the driver load" test. Not "does it report memory errors correctly".

-Tony

Re: EDAC support for 'Rocket Lake'

[Christian Lorenz]

Am 06.10.22 um 19:03 schrieb Luck, Tony:
>> I've really only added hardware I have access to, and we don't have
>> 'Rocket Lake'. I'm happy to help work on support if you have a datasheet
>> and can test. It may be as easy as adding some new pci ids for example:
> 
> Jason,
> 
> Some web searching hasn't turned up a public data sheet. The EDS volume 2
> is Intel Doc # 615924. Do you have access through Akamai? If that confirms that
> you just need to add some new PCI IDs etc. then I can get someone to make
> it OK to post a patch.
> 
> Testing is harder. I can probably get access to a rocket lake internally. But
> I don't think those support ACPI error injection. So would only be able to basic
> "does the driver load" test. Not "does it report memory errors correctly".
> 
> -Tony

Hi,

I can try to organize a piece of HW.
Let me know if that would help, and who and where this HW would be needed.
I will check with our HW vendor and we will see if things can be made possible.

Christian

Re: EDAC support for 'Rocket Lake'

[Christian Lorenz]

Am 21.10.22 um 14:00 schrieb Christian Lorenz:
> 
> Am 06.10.22 um 19:03 schrieb Luck, Tony:
>>> I've really only added hardware I have access to, and we don't have
>>> 'Rocket Lake'. I'm happy to help work on support if you have a datasheet
>>> and can test. It may be as easy as adding some new pci ids for example:
>>
>> Jason,
>>
>> Some web searching hasn't turned up a public data sheet. The EDS volume 2
>> is Intel Doc # 615924. Do you have access through Akamai? If that confirms that
>> you just need to add some new PCI IDs etc. then I can get someone to make
>> it OK to post a patch.
>>
>> Testing is harder. I can probably get access to a rocket lake internally. But
>> I don't think those support ACPI error injection. So would only be able to basic
>> "does the driver load" test. Not "does it report memory errors correctly".
>>
>> -Tony
> 
> Hi,
> 
> I can try to organize a piece of HW.
> Let me know if that would help, and who and where this HW would be needed.
> I will check with our HW vendor and we will see if things can be made possible.
> 
> Christian

Hi,

as I didn't get any response, I would like to ask again.
I understood that adding/fixing the EDAC support for Rocket Lake is lacking
available HW.
Is there any demand from one of you for Rocket Lake HW?
Let me know, so that we can check with our vendor.

Thanks.

Christian