Alan Stern wrote:
> In other words, I'm guessing that you're suffering from hardware memory
> errors.  A possible way to test this is to modify the patch.  In
> td_free() where it adds the line:
>
> +			ohci_dbg(hc, "(%d %d) %p -> %p\n", hash, n, prev, *prev);
>
> instead add this code:
>
> +			barrier();
> +			ohci_dbg(hc, "(%d %d) %p -> %p [%p]\n", hash, n,
> +					prev, *prev, td->td_hash);
>
> If we find that the value of *prev differs from the value of
> td->td_hash then we'll know for certain.  (Or maybe the presence of the 
> barrier() will cause the object code to change in a way that prevents 
> the error from occurring.)
>
> Alan Stern
>   
Hmm, I applied the changes and I did not see a place where *prev differs 
from td->td_hash. I have run memtest86+ on this box and it has passed 16 
times, so I do not suspect a hardware memory error. What do you think? 
Attached is the latest dmesg output.

Sean