On 07/19/2011 09:46 AM, Andi Kleen wrote:
> On Tue, Jul 19, 2011 at 09:39:56AM -0400, Prarit Bhargava wrote:
> 
>> The DMI specification has not been updated since January of 2003.  It
>> has been replaced by SMBIOS.
>> 
>
> Yes of course, but dmidecode and the current DMI layer implements
> both anyways, don't they? (ok if you don't count the dynamic interfaces)
>
> The tables are very similar, there are just more entries in SMBIOS.
> 

That's not the way I understand it (at least from reviewing the two
different specifications).  DMI is not SMBIOS.  They are two very
different things -- we (linux kernel) have bastardized the name DMI and
really have been using the SMBIOS tables.  It is NOT a DMI implementation.

SMBIOS *CAN* contain a DMI table but saying that SMBIOS is DMI + a few
more tables is really a stretch IMO.

>   
>>     
>>>> 3.	Every other platform without DMI would benefit from the
>>>> 	interface being generic
>>>>     
>>>>         
>>> Can you expand on that? The information will be always system
>>> specific anyways. Do you really think there's that much commonality?
>>>
>>>   
>>>       
>> There seems to be some commonalities.  We have other arches checking for
>> model and vendor info.
>>     
> That's two fields out of hundreds. Does that need a common layer?
>   

> Right now I still fail to see the point of all of this.
>
> At some point I wanted a slightly more expansive sysfs interface for SMBIOS
> to avoid having to start mcelog as root for reading /dev/mem, but I don't
> think such a complicated approach is justified for that. What are 
> the other use cases?
>   

See my previous email re: type 15 structure and trying to jam it into
the existing dmi layer.

P.
> -Andi
>