Problems with EDAC coexisting with BIOS
Quoted 6 lines · Mark
Quoted 6 lines
-----Original Message-----
From: Tim Small [mailto:[email protected]]
Sent: Wednesday, May 03, 2006 1:25 PM
To: Alan Cox
Cc: Ong, Soo Keong; Gross, Mark;
Quoted 9 lines
Quoted 9 lines
LKML; Carbonari, Steven; Wang, Zhenyu Z
Subject: Re: Problems with EDAC coexisting with BIOSQuoted 4 lines · Alan Cox
Quoted 1 line · Soo Keong
To me, periodical is not a good design for error handling, it
wastes
Quoted 5 lines
Quoted 5 lines
Quoted 5 lines
Quoted 4 lines
transaction bandwidth that should be used for other more productive purposes.
Quoted 10 lines
Quoted 1 line
Quoted 1 line
The periodical choice is mostly down to the brain damaged choice of
NMI
Quoted 8 lines
Quoted 3 lines
as the viable alternative, which is as good as 'not usable'
Hi,
As I believe that the majority of the bluesmoke/EDAC developers are (were) operating under the assumption that it would be possible to do something with NMI-signalled errors, I was wondering what the
problems
Quoted 1 line
Quoted 1 line
with using NMI-signalled ECC errors were?
Dave Peterson created the NMI patch for bluesmoke that is currently operational in the bluesmoke project. When bluesmoke was submitted, the advice we received was to defer until later to apply that NMI patch after we had gotten EDAC into the kernel and more stable.
Since Dave has stepped down, that leaves me to refactor that patch for EDAC release. It is on my TODO list, but not yet risen to the top.
Once concern I had yet to research is: Do all motherboards route the NMI signal properly? Is that a function of the BIOS? etc, etc
Quoted 5 lines
Soo Keong or Steve, can you answer this one?
Quoted 2 lines
Are there some systems states in which an incoming NMI throws a
spanner
Quoted 1 line
Quoted 1 line
in to the works in an unrecoverable way? If this is the case, is it
so
Quoted 6 lines
Quoted 4 lines
on all x86/x86-64 systems, or just a subset, and is there no way to implement some sort of top half / bottom half style NMI handler cleanly? As I am certainly not an x86 architecture expert, I would appreciate any input from the resident gurus ;o).
I don't think NMI's are so much the problem as those blasted SMI's.
And
Quoted 7 lines
SMI's are not for sharing :(
These problems are BIOS specific, Your Mileage Will Vary from one bios to the next.
Quoted 2 lines
Quickly returning to the original problem - I know this isn't a
proper
Quoted 3 lines
Quoted 3 lines
API by any stretch of the imagination, and would require changes to existing BIOSs, but the EDAC module could reprogram the chipset error-signalling registers, so that an ECC error no longer triggers
an
Quoted 8 lines
Quoted 1 line
SMI. The BIOS SMI handler could then read the signalling registers,
and
Quoted 1 line
leave the ECC registers well alone if ECC errors are not set to
generate
Quoted 1 line
an SMI.
It's not the SMI from ECC events that are the problem. It's the BIOS assuming no one else would want to use those registers on Dev0:Fun1,
and
Quoted 2 lines
the fact that you still end up with a race between the OS and the BIOS SMI to handle the events.
exactly and for that reason we might have a tar baby to wash and clean up somehow
doug t