Experiences Getting SystemTap work on OLPC Machine

William Cohen wcohen at redhat.com
Tue Mar 20 09:47:56 EDT 2007


Albert Cahalan wrote:
> Dan Williams writes:
> 
>> You may want to update to the Q2B81 firmware which disables the branch
>> prediction workaround, and yields about 25% performance increase when
>> using any USB network devices (including the builtin wireless).
> 
> Uh, is that safe?
> 
> I thought that this was required to avoid causing errors in the
> CPU's branch prediction history. Are we now just taking chances?
> If USB got fixed to use uncached RAM, what about other devices?
> _______________________________________________
> Devel mailing list
> Devel at laptop.org
> http://mailman.laptop.org/mailman/listinfo/devel

Looking through the "AMD GeodeTM GX Processors Silicon Revision 2.1 
Specification Update" (31533E_gx_2.1_specupdate.pdf) found the following entry:

1.36 Segment register misuse in AC stage of CPU execution pipeline

Is this the one underlying cause. Do you have a pointer to the issue?

Virtually every processor uses branch prediction to a improve performance. 
Disabling the CPU's branch prediction hardware is a huge performance hit.

-Will



More information about the Devel mailing list