The FC6 SATA errors were due to a defect in that kernel's SATA subsystem and have nothing to do with this problem.
I've run numerous tests the past week using kernel.org 2.6.19. I've tried a SMP kernel booted with "maxcpus=1" as well as a monoproc kernel built from the same .config file with the only change being the SMP option not being selected. I've also tried with "mem=1g" and "mem=2g". I've also tried with "iommu=soft". All result in corruption as previously described.
In the past week I've logged 112 corruption events out of 892 test cycles (12.5% failure rate). Of the nine files being copied It is always one or more of files 1, 3, 4 or 5 that are corrupted. The other five files have never been corrupted. File #4 shows corruption at twice the rate of the other three:
So, yes, I think the mainboard is defective but probably not in the manner you are suggesting. As an educated guess I think there may be a design defect with either the nForce 4 chipset or the ASUS A8N mainboard that is causing bus crosstalk. But given that others have reported similar problems with mainboards from other vendors that are based on the nForce 4 chipset it seems likely the fault lies with the nForce chipset. Or perhaps the nForce specifications that mainboard designers are using.
What would you like to do? I've got a system on which I can reproduce this problem 12.5% of the time. That system won't be used for anything else until root cause is determined or we agree to give up (at which point I'll throw the mainboard in the garbage since it can't be trusted). What tests would you propose I run? What data would you like me to provide?
Using this web site is awkward. If you would prefer to email me directly to discuss an action plan please do so: firstname.lastname@example.org
. If you email me I'll reply with phone numbers where you can reach me.