Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 10 Jan 2001 10:13:50 -0600
From:      "Jason Smethers" <jsmethers@pdq.net>
To:        "Chris Dillon" <cdillon@wolves.k12.mo.us>
Cc:        "FreeBSD Chat List" <freebsd-chat@FreeBSD.org>
Subject:   Re: ECC worth the extra cost for SOHO server? 
Message-ID:  <00d301c07b20$51f42aa0$edcf1f40@pdq.net>
References:  <Pine.BSF.4.21.0101100849020.32238-100000@mail.wolves.k12.mo.us>

index | next in thread | previous in thread | raw e-mail


From: "Chris Dillon" <cdillon@wolves.k12.mo.us>
> Yes, this might be possible.  I found the datasheet on Intel's
> developer site for the 82443BX, and it mentions which chipset
> registers you need to poke at to enable and disable certain
functions.
> Now, if it were as easy as identifying a new NMI "type" in
> intr_machdep.c:isa_nmi(), as I thought it might have been, it would
> have been a simple cut-n-paste job (about as much as I've ever
done),
> but poking at the chipset registers is something I've never done
> before, and wouldn't even know where to start. :-)
>
> Incidentally, I did find out that when the chipset detects an
> uncorrectable ECC error (it corrects correctable errors in hardware,
> i.e. does "hardware scrubbing", so the OS need not take any action
> with correctable errors), it asserts SERR#, a hardware line.  How
the
> OS would know when this was asserted, or wether as a result of this
an
> NMI is triggered, I don't know.

SERR should generate an NMI on IA-32 processors. You need to figure
out how to check for an ECC error.

Looking at the data sheet you need to do a four byte PCI configuration
read of the EAP (Error Address Pointer Register) at offset 0x80. Bits
31:12 is the page that faulted, bit 1 is a multibit ECC error, and bit
0 is a single bit ECC error.

Other Intel sheets show that this is chipset dependent. One way to do
this currently may to add an ECC quirk extraction off the PCI bus that
can be called to check the chipset for an ECC error then call this
quirk routine as a last resort if there was no other explanation for
the NMI.

Looks like buying are reading all those Mindshare Inc. books is paying
off =).

- Jason



To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-chat" in the body of the message



help

Want to link to this message? Use this
URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?00d301c07b20$51f42aa0$edcf1f40>