[PATCH v2] powernv/opal: Handle OPAL_WRONG_STATE error from OPAL fails

[Vipin K Parashar]

Hi Stewart/Michael,

Thanks!! for review.

Responses as below:


On Wednesday 01 March 2017 02:38 AM, Stewart Smith wrote:
> Vipin K Parashar <vipin at linux.vnet.ibm.com> writes:
>> Added check for OPAL_WRONG_STATE error code returned from OPAL.
>> Currently Linux flashes "unexpected error" over console for this
>> error. This will avoid throwing such message and return I/O error
>> for such OPAL failures.
>>
>> Signed-off-by: Vipin K Parashar <vipin at linux.vnet.ibm.com>
>> ---
>> Changes in v2:
>>   - Added log message indicating sleeping/offline core
>>     for OPAL_WRONG_STATE
>>
>>   arch/powerpc/platforms/powernv/opal.c | 5 ++++-
>>   1 file changed, 4 insertions(+), 1 deletion(-)
>>
>> diff --git a/arch/powerpc/platforms/powernv/opal.c b/arch/powerpc/platforms/powernv/opal.c
>> index 86d9fde..8af230e 100644
>> --- a/arch/powerpc/platforms/powernv/opal.c
>> +++ b/arch/powerpc/platforms/powernv/opal.c
>> @@ -869,8 +869,11 @@ int opal_error_code(int rc)
>>   	case OPAL_UNSUPPORTED:		return -EIO;
>>   	case OPAL_HARDWARE:		return -EIO;
>>   	case OPAL_INTERNAL_ERROR:	return -EIO;
>> +	case OPAL_WRONG_STATE:
>> +		pr_notice("%s: Core sleeping/offline\n", __func__);
>> +		return -EIO;
> Since this is part of opal_error_code() though, this will be printed for
> any OPAL call that returns that.

opal_error_coder is used by functions to handle OPAL error codes
and return Linux error codes. Apart from opal_get_sensor_data ()
in opal-sensor.c, opal_error_code  is also getting invoked from
opal_get_sys_param( ) in opal-sysparam.c.

Handling OPAL_WRONG_STATE in opal_error_code itself, seems
modular  and avoids extra checks for OPAL_WRONG_STATE after
opal_error_code usage in multiple functions.

opal_error_code is already adding a message upon OPAL_WRONG_STATE
return, so its already leaving trace about Sleeping core causing XSCOM
failure. By returning OPAL_WRONG_CODE from opal_error_code are we
planning some action like on-lining back the sleeping or off-lined core ?

>
> Why not have the sensor code do this:
>
> rc = opal_sensor_read(foo)
> if (rc == OPAL_WRONG_STATE)
>     return -EIO;
> else
>     return oal_error_code(rc);
>
> ?
>
>>   	default:
>> -		pr_err("%s: unexpected OPAL error %d\n", __func__, rc);
>> +		pr_err("%s: Unexpected OPAL error %d\n", __func__, rc);
> Do we need this?
>

This print helps in alerting about OPAL return codes that aren't 
supported in
running Linux version. Helpful in catching OPAL return code that missed
out detection check in Linux.
Shall we consider reducing message severity from pr_err to pr_warn ?