NMI error handling
 

Hello experts:

I am working on documenting and, if required, implementing error handling code for a project. We have an Intel Tylersburg IO Hub on the board. According to the datasheet, the IOH will generate an NMI for any error it encounters, and then software can read some IOH registers to discern what the condition is. It is also possible to program the IOH to generate SMI instead but I don't think I want BIOS doing any of the error handling. Sounds fair enough.

What I am confused about is: it seems a little strange that I have to go into the Linux NMI handler and add a bunch of stuff for a specific device. The existing NMI handler I looked at (do_nmi() in arch/x86_64/kernel/nmi.c) was pretty simple and straightforward and didn't seem to handle any of the errors reported by my IOH. Also, how do I access the IOH registers within the NMI? Would IOH registers be at a well-known memory address? I've asked Intel if a software device driver is needed for the IOH, but they said "no, everything should be configured by BIOS." But they have indicated that some code for error handling may be needed.

Thanks for any advice!!

-Z