[PATCH v7 9/9] docs: ABI: testing: Document the Ampere Altra Family's SMpro sysfs interfaces

[Quan Nguyen]

On 21/03/2022 17:03, Greg Kroah-Hartman wrote:
> On Mon, Mar 21, 2022 at 04:46:36PM +0700, Quan Nguyen wrote:
>>
>>
>> On 21/03/2022 15:23, Greg Kroah-Hartman wrote:
>>> On Mon, Mar 21, 2022 at 03:13:55PM +0700, Quan Nguyen wrote:
>>>> Add documentation for the Ampere(R)'s Altra(R) SMpro sysfs interfaces
>>>>
>>>> Signed-off-by: Quan Nguyen <quan at os.amperecomputing.com>
>>>> ---
>>>> Changes in v7:
>>>>     + First introduce in v7     [Greg]
>>>>
>>>>    .../sysfs-bus-platform-devices-ampere-smpro   | 133 ++++++++++++++++++
>>>>    1 file changed, 133 insertions(+)
>>>>    create mode 100644 Documentation/ABI/testing/sysfs-bus-platform-devices-ampere-smpro
>>>>
>>>> diff --git a/Documentation/ABI/testing/sysfs-bus-platform-devices-ampere-smpro b/Documentation/ABI/testing/sysfs-bus-platform-devices-ampere-smpro
>>>> new file mode 100644
>>>> index 000000000000..9bfd8d6d0f71
>>>> --- /dev/null
>>>> +++ b/Documentation/ABI/testing/sysfs-bus-platform-devices-ampere-smpro
>>>> @@ -0,0 +1,133 @@
>>>> +What:		/sys/bus/platform/devices/smpro-errmon.*/errors_[core|mem|pcie|other]_[ce|ue]
>>>
>>> Please split this out as one entry per file.
>>>
>>
>> These sysfs share same format of HW errors (the 48-byte Arm vendor specific
>> HW error record) but for separate HW domains: Core, PCIe, Mem... etc
>>
>>>> +KernelVersion:	5.14
>>>
>>> 5.14 is a long time ago.
>>>
>>>> +Contact:	quan at os.amperecomputing.com
>>>> +Description:
>>>> +		(RO) Contains the 48-byte Ampere (Vendor-Specific) Error Record, see [1]
>>>> +		printed in hex format as below:
>>>> +
>>>> +		AA BB CCCC DDDDDDDD DDDDDDDDDDDDDDDD DDDDDDDDDDDDDDDD \
>>>> +		   DDDDDDDDDDDDDDDD DDDDDDDDDDDDDDDD DDDDDDDDDDDDDDDD
>>>> +		Where:
>>>> +		  AA       : Error Type
>>>> +		  BB       : Subtype
>>>> +		  CCCC     : Instance
>>>> +		  DDD...DDD: Similar to the Arm RAS standard error record
>>>
>>> No, this is not a valid sysfs file, sorry.  This should just be one
>>> value per file.
>>>
>>
>> This 48-byte value is unable to separate into smaller values because it
>> contain all information necessary to indicate a single HW error as per ARM
>> RAS supplement document [1]. The format is to make it read-able other than a
>> single 48-byte hex value.
>>
>> [1] https://developer.arm.com/documentation/ddi0587/latest/
> 
> Just export the 48 byte hex value and make userspace split it up if it
> wants to do so.  Don't do things in the kernel that can be done in
> userspace.
> 

Thanks Greg for the suggestion,
Will do this in my next version.

Thanks,
- Quan