Files corrupt on copy
Andy Campbell
fedora at starsend.force9.co.uk
Sun Jun 28 20:28:24 UTC 2009
On Sun, 28 Jun 2009 19:22:03 +0200, Roberto Ragusa wrote:
> Andy Campbell wrote:
>
>> Hmm always out by 1 ...
>>
>> [trantor] ..mp2/tmp/src $while true
>> while> do
>> while> cp file1.zip new.zip ; cmp -l file1.zip new.zip while> done
>> 107757013 271 270
>> 109383125 206 207
>> 85093653 373 372
>> 77726613 206 207
>> 85093653 373 372
>> 85899797 373 372
>> 38258517 373 372
>> 109383125 206 207
>> 107757013 271 270
>> 109383125 206 207
>> 142459477 126 127
>> 107757013 271 270
>> 40550997 171 170
>> 85093653 373 372
>> 107757013 271 270
>> 110261013 371 370
>> 62581653 371 370
>> 109383125 206 207
>> 110526037 71 70
>> 109383125 206 207
>> 40550997 171 170
>> 77726613 206 207
>> 109383125 206 207
>> 107757013 271 270
>
> The error is always in the last bit.
> And there is also a strong similarity on the final part of the position
> error, if expressed in hexadecimal.
>
> If you paste your numbers into this command:
>
> $ while read a b; do printf %08x"\n" $a; done
>
> you get this:
>
> 066c3dd5
> 06850dd5
> 05126d15
> 04a20395
> 05126d15
> 051eba15
> 0247c755
> 06850dd5
> 066c3dd5
> 06850dd5
> 087dc255
> 066c3dd5
> 026ac255
> 05126d15
> 066c3dd5
> 06927315
> 03baeb95
> 06850dd5
> 06967e55
> 06850dd5
> 026ac255
> 04a20395
> 06850dd5
> 066c3dd5
>
> The errors are all of kind: ......X5 where x=1,5,9,d.
>
> So the errors appear only at spots with a distance of 16*4=64 bytes.
>
> Strong suspicion on your hardware: CPU (defective L2 cache line?),
> chipset or memory (did you try one 4 GiB and then the other 4GiB stick?)
>
> Oh, hardware problems could also be caused by a defective power supply
> or... a motherboard not well screwed on the chassis (this one made me
> mad some years ago).
> You could also try something related to power/speed management.
>
> The only alternative is a kernel bug, but as it is touching only one bit
> it is not likely.
>
> Best regards.
> --
> Roberto Ragusa mail at robertoragusa.it
Thanks
The memory is 4*2Gb sticks - I've tried each pair individual. Also
left memtest running with full 8gb today - passed 4 times.
Also don't seem to have problems using /tmp which is mounted as tmpfs, so
don't think its memory, or CPU - seems to be hardware related directly to
writing to physical disks.
I'll pull it all apart and rebuild it in the week - I have a spare
motherboard I can swap in. Hopefully not the power supply as it was
an overkill, never have to worry about PSU purchase of a 1000W Enermax
Galaxy.
More information about the fedora-list
mailing list