Robert Davies wrote:
[Charset iso-8859-1 unsupported, filtering to ASCII...]
> > How's this?
> >
> > /* Here is what the APIC error bits mean:
> > 0x01 0: Send CS error
> > 0x02 1: Receive CS error
> > 0x04 2: Send accept error
> > 0x08 3: Receive accept error
> > 0x10 4: Reserved
> > 0x20 5: Send illegal vector
> > 0x40 6: Received illegal vector
> > 0x80 7: Illegal register address
> > */
> Thanks for the clear info Rogier, the lkml FAQ describes this APIC
> CRC as being weak. Have you any idea if these CPU comms APICs, may
> result in data corruption?
The CRC, as far as I know, is 8-bit. That means that on average,
a "broken" message will be accepted as valid once in every 256 times
that 'something goes wrong'.
Now if that "something goes wrong" can be proven to be solely due to
random alpha particles (i.e completely random), and that only 1 in
10000 messages is corrupted by such an event, then statistically it is
VERY unlikely that two "bit-errors" end up in the same message. And an
8 bit CRC can protect against single bit errors with 100% certainty.
However, the fact that the BP6 is very prone to these errors, and
other boards aren't, means that something else is going on. Suppose
one in 10000 messages is "completely garbled", then we suddenly do
have 1 in 2560000 messages which gets corrupted and accepted as
"passed the CRC".
When bad messages get accepted as valid, odd stuff may happen. Like
lockups.
It is NOT that "your data" is communicated over the APIC line. So it
should not directly corrupt your data.
Roger.
--
** [EMAIL PROTECTED] ** http://www.BitWizard.nl/ ** +31-15-2137555 **
*-- BitWizard writes Linux device drivers for any device you may have! --*
* There are old pilots, and there are bold pilots.
* There are also old, bald pilots.
--
=- To unsubscribe, email [EMAIL PROTECTED] with the -=
=- body of "unsubscribe linux-abit". -=
