Date: Thu, 13 Apr 1995 22:25:08 +1596657 (MET DST) From: Wilko Bulte <wilko@yedi.iaf.nl> To: FreeBSD-hackers@FreeBSD.org (FreeBSD hackers list) Subject: NMI/parity error reporting Message-ID: <199504132025.WAA00806@yedi.iaf.nl>
next in thread | raw e-mail | index | archive | help
Sorry if this is a duplicate. I had the impression it didn't got to
-hackers.
-----
Currently (== 1.1.5 & 2.0R) the kernel handles NMI parity interrupts by
displaying an error and panicing. I assume -current does something
similar.
What about the following:
isa_nmi:
check_if_actually_parity_int
if not, report and panic
mask NMI for a while
clear parity hardware
install special NMI catcher
enable NMI
loop thru physmem from 0 to top of memory, scanning
memory, hoping error re-occurs
if it does, have catcher return *approximate* address
of error
if not, display message telling user
finally PANIC!
There are caveats (e.g. caches?!) but would the above be something that
works?
I know ATT V.3 did do address reporting. Lacking decent hardware (address
register containing the error address, like a decent PDP-11 ;-) I see no
other way to do it.
Reason for asking: I have a mainboard giving me *1* error, on boot. After
that, it runs solid for days.
I'd be happy to try an approach like outlined above, but it probably needs
assembly code support etc. Intel assembly is something I'm not familiar with
(and actually, would rather not _become_ familiar with ;-).
Comments and suggestions are welcome
Wilko
_ __________________________________________________________________________
| / o / / _ Wilko Bulte email: wilko@yedi.iaf.nl
|/|/ / / /( (_) Private FreeBSD site - Arnhem - The Netherlands
--------------------------------------------------------------------------------
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199504132025.WAA00806>