dma errors with sata_sil and Seagate disk

dma errors with sata_sil and Seagate disk

[John Lash]

I'm using the sata_sil driver on a Tyan 2885. It has onboard Silicon Image 3114
SATA interfaces. There's one disk connected now, a Seagate ST380013AS. I've been
seeing persistent problems with errors (DriveStatusError) on the device. It
seems to happen pretty quickly under a load (trying to copy a couple of GB of
files to the disk).

I noticed that the sil_blacklist had quite a few Seagate entries and on
searching the web, found that these were dma problems that looked pretty similar
to mine. Since then I've added this model number to the sil_blacklist table,
rebuilt and rebooted. Now it's stable (just copied 20 GB of data to it). Is this
another problematic drive? Or am I masking the original problem?

I see the problem consistently on 2.6.8.1, 2.6.9, and 2.6.10-rc2-bk13. Also saw
the problems when I tried this on an older AMD MP system with a SI 3112 PCI
controller card.

I don't see any indication that Seagate has released any public firmware
upgrades for this drive. Anybody have a suggestion?

--john

Re: dma errors with sata_sil and Seagate disk

[Alan Cox]

On Mer, 2004-12-01 at 17:50, John Lash wrote:
> I don't see any indication that Seagate has released any public firmware
> upgrades for this drive. Anybody have a suggestion?

Don't mix seagate drives and SI311x hardware is the best suggestion.
Even if you activate the workaround for the problem you take a
performance hit.

Please send Jeff Garzik a patch for the the change you made of course.

Re: dma errors with sata_sil and Seagate disk

[John Lash]

On Wed, 01 Dec 2004 23:41:22 +0000
Alan Cox <alan@lxorguk.ukuu.org.uk> wrote:

> On Mer, 2004-12-01 at 17:50, John Lash wrote:
> > I don't see any indication that Seagate has released any public firmware
> > upgrades for this drive. Anybody have a suggestion?
> 
> Don't mix seagate drives and SI311x hardware is the best suggestion.
> Even if you activate the workaround for the problem you take a
> performance hit.
> 
> Please send Jeff Garzik a patch for the the change you made of course.
> 
> 

Thanks Alan, that's pretty clear ;-) Yes, performance hit, big time, 9-10 MB/sec, load average
through the roof on a big copy. Pretty much what I was expecting though. I'll look around work and
see if there is anybody I can trade with. Odd that those two devices are so wildly incompatible.

I'll roll up a patch for the change in the morning.

--john

Re: dma errors with sata_sil and Seagate disk

[Måns Rullgård]

Alan Cox <alan@lxorguk.ukuu.org.uk> writes:

> On Mer, 2004-12-01 at 17:50, John Lash wrote:
>> I don't see any indication that Seagate has released any public firmware
>> upgrades for this drive. Anybody have a suggestion?
>
> Don't mix seagate drives and SI311x hardware is the best suggestion.

Is there some problem with Seagate drives in general?  I'm using two
ST3160827AS drives on an SI3114 controller, and haven't seen any
glitches yet.  That model is not in the blacklist, and performance is
what I'd usually expect.  Is it pure luck that has kept me away from
problems?

-- 
Måns Rullgård
mru@inprovide.com

Re: dma errors with sata_sil and Seagate disk

[Alan Cox]

On Iau, 2004-12-02 at 10:01, Måns Rullgård wrote:
> Is there some problem with Seagate drives in general?  I'm using two
> ST3160827AS drives on an SI3114 controller, and haven't seen any
> glitches yet.  That model is not in the blacklist, and performance is
> what I'd usually expect.  Is it pure luck that has kept me away from
> problems?

I've never been able to get a non NDA list of the affected drives. Got
to love vendors some days

Re: dma errors with sata_sil and Seagate disk

[Måns Rullgård]

Alan Cox <alan@lxorguk.ukuu.org.ak> writes:

> On Iau, 2004-12-02 at 10:01, Måns Rullgård wrote:
>> Is there some problem with Seagate drives in general?  I'm using two
>> ST3160827AS drives on an SI3114 controller, and haven't seen any
>> glitches yet.  That model is not in the blacklist, and performance is
>> what I'd usually expect.  Is it pure luck that has kept me away from
>> problems?
>
> I've never been able to get a non NDA list of the affected drives. Got
> to love vendors some days

Does this mean it is the drives which are faulty, not the controller?
These drives are both new, so I suppose known problems might have been
fixed.  FWIW, they are reported by the kernel thusly:

  Vendor: ATA       Model: ST3160827AS       Rev: 3.03
  Type:   Direct-Access                      ANSI SCSI revision: 05
  Vendor: ATA       Model: ST3160827AS       Rev: 3.00
  Type:   Direct-Access                      ANSI SCSI revision: 05

-- 
Måns Rullgård
mru@inprovide.com

Re: dma errors with sata_sil and Seagate disk

[Jeff Garzik]

Måns Rullgård wrote:
> Does this mean it is the drives which are faulty, not the controller?
> These drives are both new, so I suppose known problems might have been
> fixed.  FWIW, they are reported by the kernel thusly:


The issue is that the SiI 311x controllers send out packets (called 
"FIS's" in SATA-land) that are not multiples of 512 bytes.

This is perfectly legal according to SATA spec, but early drive 
firmwares (notoriously Seagate in this case) were written with the 
assumption that the FIS's would be multiples of 512 bytes.

	Jeff

Re: dma errors with sata_sil and Seagate disk

[John Lash]

On Fri, 03 Dec 2004 02:14:27 +0100
Måns Rullgård <mru@inprovide.com> wrote:

> Alan Cox <alan@lxorguk.ukuu.org.ak> writes:
> 
> > On Iau, 2004-12-02 at 10:01, Måns Rullgård wrote:

> Does this mean it is the drives which are faulty, not the controller?
> These drives are both new, so I suppose known problems might have been
> fixed.  FWIW, they are reported by the kernel thusly:
> 
>   Vendor: ATA       Model: ST3160827AS       Rev: 3.03
>   Type:   Direct-Access                      ANSI SCSI revision: 05
>   Vendor: ATA       Model: ST3160827AS       Rev: 3.00
>   Type:   Direct-Access                      ANSI SCSI revision: 05
> 

I know that the disk that I'm using has been sitting on a shelf since about last June. It still
seems to be a shipping model though. It's showing itself as model: ST380013AS, rev: 3.18. 

It would be nice if the problem gets fixed someday. Be even nicer if there was a firmware update
that could be applied to the older drives.....

--john

Re: dma errors with sata_sil and Seagate disk

[Jeff Garzik]

Alan Cox wrote:
> On Iau, 2004-12-02 at 10:01, Måns Rullgård wrote:
> 
>>Is there some problem with Seagate drives in general?  I'm using two
>>ST3160827AS drives on an SI3114 controller, and haven't seen any
>>glitches yet.  That model is not in the blacklist, and performance is
>>what I'd usually expect.  Is it pure luck that has kept me away from
>>problems?
> 
> 
> I've never been able to get a non NDA list of the affected drives. Got
> to love vendors some days

I seriously doubt a complete list exists, NDA or no.  You'd have to poll 
each vendor.

I also suspect that a few of the more recent Seagate additions are 
simply masking a problem in the BIOS.

SiI 311x problems have a history of resolving themselves through BIOS 
updates and tweaks.  Most recently a lockup was solved by tweaking a 
'byte enable' setting in an nForce mobo BIOS.

	Jeff

Re: dma errors with sata_sil and Seagate disk

[raz0]

Hi,

I was wondering whether you know how this bug is dealt with in the siI Windows
drivers. Are they also using some fix that limits the performance? I remember
my friend pulling a file via a cross-over cable (ie: no switch) with some
26megs/s, so I guess not? Would you be so kind to explain this to me? :)

I'm running 2.6.10-rc3 with the blacklisted ST3120026A on the SiI 3512A
controller. When I try 'hdparm -t /dev/sda', I get this:

raz0:/home/morten# hdparm -t /dev/sda

/dev/sda:
 Timing buffered disk reads:   46 MB in  3.08 seconds =  14.94 MB/sec
HDIO_DRIVE_CMD(null) (wait for flush complete) failed: Inappropriate ioctl
for device

Is this the same bug? I haven't noticed any problems other than the hdparm
one.

Please CC me if you reply; I am not on the list.

	Morten

>> I've never been able to get a non NDA list of the affected drives. Got
>> to love vendors some days
>
>I seriously doubt a complete list exists, NDA or no.  You'd have to poll

>each vendor.
>
>I also suspect that a few of the more recent Seagate additions are 
>simply masking a problem in the BIOS.
>
>SiI 311x problems have a history of resolving themselves through BIOS 
>updates and tweaks.  Most recently a lockup was solved by tweaking a 
>'byte enable' setting in an nForce mobo BIOS.
>
>	Jeff