[CentOS] SATA errors in log

[m.roth at 5-cent.us]

Steve Brooks wrote:
> On Fri, 22 Jun 2012, m.roth at 5-cent.us wrote:
>> Steve Brooks wrote:
>>>
>>> I have a SATA PCIe 6Gbps 4 port controller card made by Startech. The
>>> kernel (Linux viz1 2.6.32-220.4.1.el6.x86_64) sees it as
>>>
>>>   Marvell Technology Group Ltd. 88SE9123
>>>
>>> I use it to provide extra SATA ports to a raid system.
>>> The HD's are all "WD2003FYYS" and so run at 3Gbps on the 6Gbps
>>> controller. However I am seeing lots of instances of errors like this
>>>
>>> Jun 22 03:13:23 viz1 kernel: ata13.00: exception Emask 0x10 SAct 0x4
>>> SErr
>>> 0x400000 action 0x6 frozen
>>> Jun 22 03:13:23 viz1 kernel: ata13.00: irq_stat 0x08000000, interface
>>> fatal error
>>> Jun 22 03:13:23 viz1 kernel: ata13: SError: { Handshk }
>>> Jun 22 03:13:23 viz1 kernel: ata13.00: failed command: WRITE FPDMA
>>> QUEUED
>>> Jun 22 03:13:23 viz1 kernel: ata13.00: cmd
>>> 61/e8:10:98:05:1b/01:00:66:00:00/40 tag 2 ncq 249856 out
>>> Jun 22 03:13:23 viz1 kernel: ata13.00: status: { DRDY }
>>> Jun 22 03:13:23 viz1 kernel: ata13: hard resetting link
>> <snip>
>> Crap. First question: what make & model are the drives on it? If they're
>> Caviar Green, you're hosed. WD, and *maybe* Seagate as well, disabled a
>> certain function you used to be able to set on the lower cost,
>> consumer-grade models (in '09, I believe), and so when a server
>> controller is trying to do i/o, and has a problem, in server-grade drives,
>>  it gives up after something like 6 sec, and does error handling, I *
>> think* to other sectors. The consumer ones, on the other hand, keep trying
>> for 1? 2? *minutes*; the disabled function allowed a used to tell it to
>> give up in a shorter time. Meanwhile, a hardware controller will, as I
said,
>> have fits.
>>
>>        mark "you'd think I just spent months dealing with this...."
>>
>
> As mentioned in the original post the drives are all "WD2003FYYS". I am

Missed the original post; sorry.

> convinced it has nothing to do with TLER enabled on the WD drives as we

Thanks, that was the acronym I was trying to remember.

> run hundreds of them using linux mdadm raid on motherboard SATA
> controllers with no problems in the last eight or so years. This appears
> to be specific to the SATA PCIe 6Gbps 4 port controller card made by
> Startech. There are four other HD's (WD2003FYYS) in the machine running on
> an onboard "Intel Corporation Patsburg 6-Port SATA AHCI Controller" with
> no problems.

I also see those are "enterprise" drives, not consumer grade, which
implies that they ought to work. It still looks to me as though it's
timing out, which I'd think is a function of the RAID card. You might see
if it has any firmware configuration options.

       mark