We've managed to build a binary which crashes both our alphas with
100% consistency.
The crash looks similar to one reported recently on axp-list by Shane
DeRidder <shane@together.net>. The console starts spewing out lines
like:
0 0000000000000198 0000019800000110
10 0000000100000084 00000001201dc850
20 000006f800000000 ffffffffffffffff
30 ffffffffffffffff 0000015555cac5b0
40 0000000120535690 00000155557c1db8
50 0000000120534ed2 0000000000000008
...
(the numbers are made up because they fly by too fast to record, and
C-S doesn't stop them).
The machine responds to pings. The keyboard isn't dead (the capslock
key toggles the keyboard capslock light), but the machine doesn't
respond to control-alt-delete, can't be telnetted into, won't switch
virtual consoles, needs to be reset, ...
The binary was made by rebuilding f2c & libf2c from the latest netlib
source & compiling the following Fortran program:
program test
c-----------------------------------------------------------------------
integer i
integer*2 i2
real*4 r4
real*8 r8
c-----------------------------------------------------------------------
do i = 1, 10
i2 = i
r4 = i
r8 = i
write(6,*) i,i2,r4,r8
enddo
end
If anyone would like to work on this problem, I can email you the
statically linked binary which does the job.
If anyone has any ideas of how to fix this, please tell me.
This is with kernel 2.0.30, as supplied by Aspen Systems - I presume
they patched the virgin 2.0.30 with gatekeeper's alpha patch, but I
don't know for sure.
Thanks,
-- Harvey J. Stein Berger Financial Research hjstein@bfr.co.il