[LinuxBIOS] Intel refactoring and microcode updates

[joe at smittys.pointclark.net]

Quoting Corey Osgood <corey.osgood at gmail.com>:

> Carl-Daniel Hailfinger wrote:
>> On 11.01.2008 02:33, Corey Osgood wrote:
>>
>>> joe at smittys.pointclark.net wrote:
>>>
>>>
>>>> Quoting Corey Osgood <corey.osgood at gmail.com>:
>>>>
>>>>
>>>>> Do you mean the microcode files? If so, the microcode update looks like
>>>>> this:
>>>>>
>>>>> Header
>>>>> Update Revision
>>>>> Date
>>>>> Processor Signature (CPU ID)
>>>>> ...
>>>>>
>>>>> So, the 4th entry in the update is always the CPU ID, and conveniently
>>>>> it's always the last one on the first line. It also makes grepping for
>>>>> them very easy, once you have the update broken down into smaller files.
>>>>> This is documented *somewhere* in LB, but I can't find it at the moment.
>>>>> It's also in the Intel architecture manual, volume 3a, table 9-6.
>>>>>
>>>>> In the past we labeled some CPU IDs as to what CPUs they belonged to. In
>>>>> truth, Intel uses the same CPU IDs for a variety of CPUs, for instance
>>>>> in some cases Celeron, Pentium X, and Xeons all share a common ID, since
>>>>> the core is still the same. So we can't really do that any more ;)
>>>>>
>>>>>
>>>>>
>>>> Oh ok, that makes sense.
>>>>
>>>> Acked-by: Joseph Smith <joe at smittys.pointclark.net>
>>>>
>>>>
>>> Thanks, Joe. Anyone else have anything to say? Honestly expected more
>>> feedback, but if there are no objections I'll commit it tomorrow.
>>>
>>> The other thing I forgot to mention was that all the data on CPU IDs
>>> came from the existing code and this site:
>>> http://processorfinder.intel.com. Some of them are a bit unclear on what
>>> sockets they use, but if anything comes up wrong, we can easily correct it.
>>>
>>>
>>
>> Do you see any way to solve the "size problem" for sockets with too many
>> different cores? It would also be interesting to find out if your work
>> on stripping duplicate contents gives us new opportunities to reduce
>> size even further.
>>
>> Regards,
>> Carl-Daniel
>
> I'm thinking lzma compression, it knocks the files down to about 1/3
> their current size or smaller. But I hate to introduce lzma as a
> requirement, especially just for this one task, and some distros don't
> have lzma prepackaged. I have already removed all the duplicate updates,
> so the only option is possibly eliminating more cores, for socket 604,
> and for lga775 it's either breaking things down into smaller subsets
> (pentium vs. core, as i mentioned before), or larger flash chips.
>
> joe at smittys.pointclark.net wrote:
>> Not sure what you mean? How many different cores could you put in even
>> the most popular socket, three?
>>
>> Thanks - Joe
>
> LGA775 currently has 19 (and may have more I don't know about). Using
> some rough math, I get a rounded-down size of 186KB (really is quite a
> bit more) for its updates, and that would go into both normal and
> fallback images.
>
> -Corey
>
Wow I guess I didn't realize LGA775 had so many.....


Thanks - Joe