Data corruption with sata_sil (Sil 3112)

Data corruption with sata_sil (Sil 3112)

[Simon Kirby]

Hello,

I've been having problems with Sil 3112 cards I purchased for additional
SATA ports resulting in read data corruption, about 3-5 instances over
2 GB of data, 100% reproducible.

I tried fiddling with PCI latency timer, underclocking, all sorts of
stuff, all no help.

I just rebuilt the entire box with the remains of another (went from
A7V8X (VIA) to A7N8X (NVidia), new CPU, new RAM, new power supply),
thinking the problem was related to the motherboard.  The issue followed
to the new box.

This new motherboard has an onboard Sil 3112 as well.  The old onboard
was VIA SATA, which did not corrupt anything.  The Sil 3112 onboard now
does too.

I noticed the driver has a request-limit quirk.  I tried forcing this on
-- same results.  I noticed there is a DMA-limiting quirk -- forced on,
same result.

I can't seem to find a way to make this chipset not corrupt reads.

Am I alone?

Scipt used to md5sum to find corruption:

	find $* -type f -print0 | sort -z | xargs -0 md5sum

Reproduced on any recent 2.6 kernel version (2.6.18 - 2.6.21).

Simon-

Re: Data corruption with sata_sil (Sil 3112)

[Jim Paris]

> I've been having problems with Sil 3112 cards I purchased for additional
> SATA ports resulting in read data corruption, about 3-5 instances over
> 2 GB of data, 100% reproducible.
..
> I just rebuilt the entire box with the remains of another (went from
> A7V8X (VIA) to A7N8X (NVidia), new CPU, new RAM, new power supply),
> thinking the problem was related to the motherboard.  The issue followed
> to the new box.

Have you tried different disks?  I recently spent a long time trying
to track down the same sort of problem and it ended up being a bad
HD (not a media failure, so SMART didn't report it).

> This new motherboard has an onboard Sil 3112 as well.  The old onboard
> was VIA SATA, which did not corrupt anything.  The Sil 3112 onboard now
> does too.

Maybe the VIA controller was only 1.5 Gbps and your 3112 controllers
are running at 3.0 Gbps?  Some drives have a jumper that lets you
limit their operation to 1.5, which you could try.

> Scipt used to md5sum to find corruption:
> 
> 	find $* -type f -print0 | sort -z | xargs -0 md5sum

Can you figure out the nature of the corruption?  Flipped bit, entire
blocks corrupted, etc?  Maybe make two big identical files and use
"cmp -l" to see how they read differently.

-jim

Re: Data corruption with sata_sil (Sil 3112)

[Tejun Heo]

Hello, Simon, Jim.

Jim Paris wrote:
>> I've been having problems with Sil 3112 cards I purchased for additional
>> SATA ports resulting in read data corruption, about 3-5 instances over
>> 2 GB of data, 100% reproducible.
> ..
>> I just rebuilt the entire box with the remains of another (went from
>> A7V8X (VIA) to A7N8X (NVidia), new CPU, new RAM, new power supply),
>> thinking the problem was related to the motherboard.  The issue followed
>> to the new box.
> 
> Have you tried different disks?  I recently spent a long time trying
> to track down the same sort of problem and it ended up being a bad
> HD (not a media failure, so SMART didn't report it).

Hmm... that's interesting.

>> This new motherboard has an onboard Sil 3112 as well.  The old onboard
>> was VIA SATA, which did not corrupt anything.  The Sil 3112 onboard now
>> does too.
> 
> Maybe the VIA controller was only 1.5 Gbps and your 3112 controllers
> are running at 3.0 Gbps?  Some drives have a jumper that lets you
> limit their operation to 1.5, which you could try.

3112 doesn't to 3.0 Gbps and SATA auto-negotiates transfer speed when
PHY goes online.  The jumper helps detection on some dump controllers
but shouldn't cause data corruption.

>> Scipt used to md5sum to find corruption:
>>
>> 	find $* -type f -print0 | sort -z | xargs -0 md5sum
> 
> Can you figure out the nature of the corruption?  Flipped bit, entire
> blocks corrupted, etc?  Maybe make two big identical files and use
> "cmp -l" to see how they read differently.

Yeap, please.

-- 
tejun