Random reboot question

[Tom Sylla]

On 10/17/07, John Watlington <wad at laptop.org> wrote:
>
> Is it in any way likely that memory errors could lead an x86 system
> to reboot ?

I think it is possible. If reads of PDEs or PTEs got corrupted, the
system could go down the triple fault path. Those fetches happen so
rarely that it would seem that you would see invalid opcode of other
faults reported by the kernel much more often than triple faults,
though.

Software can really only reboot two ways: write the bit in 5536 to
cause the reset, or cause a triple fault (which causes a shutdown
special cycle, which will cause the DIVIL to reset the system). Random
reboots are probably not the first one, so you are left with triple
faults.

If you want to diagnose if it is the CPU or not, disable the
reset-on-shutdown-special-cycle behavior (clear the top bit in
DIVIL_LEG_IO). If the system still resets, it wasn't a triple fault
that did it. If the system wedges instead of resets, you know it
probably was a triple fault.