I am presently in the midst of configuring an Alpha (EB164) server
running Linux (RedHat 3.0.3 - upgraded to 2.0.8) for a client. When I
arrived at my client's site this morning I was greeted with this message:
Unable to handle kernel paging request at virtual address fbfffc0004fd6d80
syslogd(55): Oops 0
pc=[<fffffc000033a820>] ps=0000
rp=[<fffffc000038d1a4>] sp=fffffc0007aa760
r0=fbfffc0004fd6d80
r1=fffffc0000000000
r2=fffffc0000471090
r3=fffffc000044c978
r8=10 r16=802 r17=sccd r18=400 r19=0 r24=24cf r25=3ffd
r26=fffffc000038d1a4
r27=fffffc0000012678
r28=1
r29=fffffc0000466080
r30=fffffc0007aa7e60
Code: 47ff041f 47ff041f 47ff041f <a4200000> 402b0521 f4200019
40011402 4440d101 a4210000
The /var/log/messages entries for the time it went down are:
Jul 25 18:27:33 alphabits kernel: eth0: mode is 100Mb/s.
Jul 25 18:27:36 alphabits kernel: eth0: mode is 10Mb/s.
Jul 26 12:59:10 alphabits syslogd 1.3-0: restart.
Jul 26 12:59:10 alphabits kernel: klogd 1.3-0, log source = /proc/kmsg started.
Jul 26 12:59:10 alphabits kernel: Error in symbol table input.
Jul 26 12:59:10 alphabits kernel: Cannot find map file.
Jul 26 12:59:10 alphabits kernel: Command line: bootdevice= bootfile= root=/dev/sda2
Jul 26 12:59:10 alphabits kernel: Alpha PCI BIOS32 revision 0.03
Jul 26 12:59:10 alphabits kernel: Probing PCI hardware.
There doesn't seem to be anything unusual there. The 100Mb/s -
10Mb/s thing happens all the time. Stopping it is my next project.
Before leaving for the day yesterday I started the 'Byte Unix
Benchmarks' running on both a Pentium and the Alpha to compare them. This
is very likely the cause of the crash.
The output from the benchmark run is this:
Dhrystone 2 without register variables 1 2 3 4 5 6
awk: /usr/local/src/bm/pgms/loops.awk:38: (FILENAME=- FNR=4)
fatal: division by zero attempted
Dhrystone 2 using register variables 1 2 3 4 5 6
awk: /usr/local/src/bm/pgms/loops.awk:38: (FILENAME=- FNR=4)
fatal: division by zero attempted
Arithmetic Test (type = arithoh) 1 2 3 4 5 6
awk: /usr/local/src/bm/pgms/loops.awk:38: (FILENAME=- FNR=4)
fatal: division by zero attempted
Arithmetic Test (type = register) 1 2 3 4 5 6
awk: /usr/local/src/bm/pgms/loops.awk:38: (FILENAME=- FNR=4)
fatal: division by zero attempted
Arithmetic Test (type = short) 1 2 3 4 5 6
awk: /usr/local/src/bm/pgms/loops.awk:38: (FILENAME=- FNR=4)
fatal: division by zero attempted
Arithmetic Test (type = int) 1 2 3 4 5 6
awk: /usr/local/src/bm/pgms/loops.awk:38: (FILENAME=- FNR=4)
fatal: division by zero attempted
Arithmetic Test (type = long) 1 2 3 4 5 6
awk: /usr/local/src/bm/pgms/loops.awk:38: (FILENAME=- FNR=4)
fatal: division by zero attempted
Arithmetic Test (type = float) 1 2 3 4 5 6
awk: /usr/local/src/bm/pgms/loops.awk:38: (FILENAME=- FNR=4)
fatal: division by zero attempted
Arithmetic Test (type = double) 1 2 3 4 5 6
awk: /usr/local/src/bm/pgms/loops.awk:38: (FILENAME=- FNR=4)
fatal: division by zero attempted
System Call Overhead Test 1 2 3 4 5 6
awk: /usr/local/src/bm/pgms/loops.awk:38: (FILENAME=- FNR=4)
fatal: division by zero attempted
Pipe Throughput Test 1 2 3 4 5 6
awk: /usr/local/src/bm/pgms/loops.awk:38: (FILENAME=- FNR=4)
fatal: division by zero attempted
Pipe-based Context Switching Test 1 2 3 4 5 6
Process Creation Test 1 2 3 4 5 6
awk: /usr/local/src/bm/pgms/loops.awk:38: (FILENAME=- FNR=4)
fatal: division by zero attempted
Execl Throughput Test 1 2 3 4 5 6
awk: /usr/local/src/bm/pgms/loops.awk:38: (FILENAME=- FNR=4)
fatal: division by zero attempted
Filesystem Throughput Test (10 second test) 1
So I guess the Filesystem Throughput Test got the better of the
system.
If you'd like any more info, just ask.
BTW - Anyone know of a benchmark package that doesn't crash the system? :)
Scott D. Webster I had far rather walk, as I do, in daily terror
Senior, Computer Science of eternity, than feel that this was only a
William Paterson College children's game in which all of the contestants
http://www.wilpaterson.edu would get equally worthless prizes in the end.
/home/students/webster/index.htm - T. S. Elliot