AMD Linux Kernel Patch Source Code Posted

H

HardOCP News

[H] News
Joined
Dec 31, 1969
Messages
0
For those of you following the whole AMD TLB erratum 298 issue, AMD has posted information on a workaround as well as the source code for the Linux 64-bit patch on the x86-64.org. According to the post from AMD, the information posted is for “demonstration purposes and is NOT being recommended to be applied upstream.”

AMD Family 10h revision B2 processors suffer from an issue in the processor TLB known as erratum 298. Erratum 298 is documented in a forthcoming update to the Revision Guide for AMD Family 10h Processors (PID 41322). The workaround in the Revision Guide document is intended to be applied by BIOS. The BIOS workaround has performance implications which can be avoided by having the OS directly workaround the issue. A Linux 64-bit patch was developed for 2.6.23.8 by AMD's OSRC team and will be posted to this list by Joerg Roedel. The patch is for demonstration purposes and is NOT being recommended to be applied upstream.
Click to expand...
 
There was a problem? I have been reading the news here every day and never read a mention of it, or even that AMD stopped shipping most Barcelona CPUs. Maybe I just missed it.
 
:D

pxc are you being facetious ?? I am surprised that no one is talking about the NB speed issue,or the,how should I phrase this ?? 'revised' AMD TDP announcements for K10 either.At any rate,hopefully this will help those in need of a fix on Linux.
 
From the AMD forum:

Tech Report tested Phenom performance with the TLB BIOS fix enabled: http://techreport.com/articles.x/13741

It takes a 13.9-19.8% performance hit on average, depending on whether or not you count bandwidth tests.

There is another problem too. The AMD 9600/9700 samples have the IMC/NB clocked at 2GHz instead of the shipping 1.8GHz. In most benchmarks it doesn't make much of a difference, but in a couple there are significant differences of 5-8% lower performance at 1.8GHz (without the TLB patch).

It's not clear if Kyle lowered the IMC/NB speeds to the shipping 1.8GHz frequency on his sample or left it at 2GHz. Will [H] revisit this if it was benchmarked at the higher speed?
 
This can be done by os instead of bios in linux, cause every1 can make fixes for it, in windows, naaah, closed source..

Linux = quite overpower over MS there:p
 
The lesson here is for Intel as well. AMD's monolithic design is biting them in the ass big time. Although the average consumer doesn't know squat about errata issues with processors, this is one of those that is causing AMD to stop shipping Barcelona's and Phenom's at higher clocks until this can be fixed.
 
jbzx86 said:
The lesson here is for Intel as well. AMD's rushed validation and unrealistic release schedule is biting them in the ass big time.
Click to expand...
Fixed.

Despite what AMD has been saying for the last year or so, Barcelona scheduled to be a 2008 product. Rushing it in a panic made this disaster.
 
You must log in or register to reply here.
Back
Top