Secret Debug registers in AMD processors
Monday, 15 November 2010

Hidden password-protected debug registers in the Athlon XP is enough to make any hardware-oriented programmer more paranoid about what could be lurking between the registers.

Banner

If you work with assembler you slowly grow to understand the architecture of the machine and to know its registers and operations as if they were your own. So it comes as a shock to discover that there is something hidden that the manufacturers built in but didn't bother to tell you about. It is even worse when you discover that they went to active lengths to stop you finding the extra hardware by password-protecting the access. Apparently this is exactly what AMD has done with its processors from the Athlon XP on. 

 

athlon

 

A reverse engineer going by the name Czernobyl (Czerno for short) has discovered that the Athlon XP series has included firmware-based debugging features that go well beyond the standard facilities. Four new register each password protected are involved and you can only gain access to them if the EDI register holds 9C5A203A otherwise an exception results.

The facilities provided are still under investigation but the registers seem to implement a conditional hardware breakpoint. You can arrange a breakpoint to occur on a block of memory specified by an address  mask and data mask. Even this facility would make reverse engineering of some types of system much easier. More probably remains to be discovered.

A really intriguing question is why AMD included this facility and then decided not to only make it public but actively to keep it secret by password protection. What does it think it was protecting?

Such discoveries, and proposals by Intel for example to include additional hardware that can be unlocked for a fee, increase the paranoia of the average hardware-oriented programmer. What else could there be hiding between the registers?

More info from Czernobyl's post on RCE.

Banner


pg_disatch - Run SQL Queries Asynchronously On PostgreSQL
24/06/2025

pg_disatch is meant to be a TLE-compliant alternative to pg_later but built on top of pg_cron. What makes it different?



Deno Not Giving Up Over JavaScript Trademark
01/07/2025

Deno has faced a setback in its attempt to get Oracle to relinquish the JavaScript Trademark. The US Patents Office Trademark Trial and Appeal Board (TTAB) dismissed Deno's fraud claim which is one th [ ... ]


More News

Last Updated ( Monday, 15 November 2010 )