[OLPC-devel] Re: Problem after installing to NAND

[Tom Sylla]

Jim Gettys wrote:
> If I knew what the PRS was, and had a copy of it, I might run it
> periodically ;-).

"Preferred Register Settings"

It is a mechanism the geode group has used for a very long time to make 
sure registers get set properly, and stay that way. After tapeout, 
before silicon, a system engineer gets all of the proper register 
settings from the silicon team. They go into a text file with a defined 
syntax. A parser runs in DOS that reads all those registers and compares 
to the text file. Differences are reported as errors. As the silicon is 
turned on, the PRS evolves, gets updated, etc. The PRS check is run at 
every validated internal BIOS release. If the BIOS makes some change 
that messes up a register, it is caught easily and quickly. The PRS 
includes MSRs, I/O space, MMIO, and PCI config space. After the silicon 
ships, BIOS partners make their BIOS ports. When they find that 
something doesn't work, and ask AMD for help with it, one of the first 
things done is to run the PRS check, to make sure they didn't screw 
something up.

Early in the LB port, there were lots of registers misconfigured or not 
set up. I made sure SteveG worked on getting them in a pretty good state 
using the PRS mechanism. There is now even a PRS checker that runs 
through FS2, so it makes it pretty easy, you could even run it on one of 
those broken "jumping to..." machines. Ron should have the final OLPC 
PRS, and the parser. You'd have to ask AMD if the parser is available to 
anyone.

Tom