[Skiboot] [PATCH v12 09/10] skiboot: Add core IMC related counter configuration

[Madhavan Srinivasan]

On Wednesday 14 June 2017 10:46 AM, Michael Neuling wrote:
>>> diff --git a/include/opal-api.h b/include/opal-api.h
>>> index 02927356d033..5dbae2d57fde 100644
>>> --- a/include/opal-api.h
>>> +++ b/include/opal-api.h
>>> @@ -1232,6 +1232,7 @@ enum {
>>>   /* Operation argument to IMC Microcode */
>>>   enum {
>>>>   	OPAL_IMC_COUNTERS_NEST = 1,
>>> +	OPAL_IMC_COUNTERS_CORE = 2,
>> How do we advertise to linux that it can do CORE vs NEST calls?
>>
>> Say we need to remove one, how does linux know which one it can call without
>> trying and getting OPAL_PARAMETER?
>>
>> If it's based on the device tree, then the patch series the wrong way around
>> since the earlier patches start adding the device tree before the calls are
>> added.  So if you apply just patches 1-7, Linux is going to blow up with a
>> bunch
>> of OPAL_PARAMETER since it'll start seeing the device tree before the OPAL
>> calls
>> have been added.
>>
>> Not a big issues for this series, but it does make me wonder.  If someone
>> updates the device tree in the PNOR to start adding a new section type for
>> IMC,
>> how do we tell linux that even though the device tree supports it, OPAL itself
>> doesn't?
>>
>> Say we have PHB counters in the future (just made up).  We add them to the IMA
>> catalog and linux, but not skiboot.  Linux is going to OPAL_PARAMETER and blow
>> up.  Linux needs to check if the device tree and skiboot have the capability.
>> Not just the device tree.
> Talking to benh offline, he suggested skiboot should parse the devicetree
> imported from the PNOR and strip out things it doesn't understand.
>
> That way Linux won't try to touch things it doesn't understand.
Yes this make sense. Nice catch mikey.

Thanks ben/mikey, will add a parser loop for the incoming
device tree from pnor to remove any unsupported "type"
of IMC devices by the OPAL.

Maddy

>
> Mikey