On 2014-12-02 19:56, Tejun Heo wrote:
> Hello,
> 
> On Fri, Oct 17, 2014 at 11:02:43PM +0100, Nuno Magalhães wrote:
>> The Toshibas would hold the OS in a RAID1 while the WDs would hold data in ZFS.
>> At the moment T1 has the OS, T2 has a partition, the WDs are empty.
>>
>> My next test will be to move T2 to the bottom of the BIOS boot
>> priority, so it'll become T1, W1, W2, T2 to see if it fails.

I've ran the above tests, the results are the same: every now and then
at least one WD fails (but is recovered if i reset it, although on
occasion i hade to issue the command more than once). Today i even had
the 2 of them fail. I attach the latest results, with the following
configuration:

T1 and T2 are Toshiba DT01ACA100
W1 and W2 are Western Digital"Red" WD10EFRX

physical SATA port , BIOS order , disk
1                  , 1          , T1
2                  , 2          , W1
3                  , 4          , T2
4                  , 3          , W2

kernel parameters: libata.force=dump_id,nohrst

Next will be to move the toshibas to physical ports 3 and 4 and try again.

> If you're still working on it, can you please try to reproduce the
> failure with the debug patch applied and report the dmesg?

What debug patch is that? I assume i'd have to get libata source from
somewhere first (end user here)? Any RTFM links i can use?

Thanks,
Nuno