Re: IRQ issues with multiple SiI3114's on Kernel 3.2

[Stan Hoeppner <stan@hardwarefreak.com>]

On 8/1/2012 6:43 PM, Stirling Westrup wrote:
> On Sun, Jul 29, 2012 at 3:00 PM, Stirling Westrup <swestrup@gmail.com> wrote:
>> On Sun, Jul 29, 2012 at 5:24 AM, Stan Hoeppner <stan@hardwarefreak.com> wrote:
>>> On 7/28/2012 6:41 PM, Stirling Westrup wrote:
>>>
>>>> Okay, it looks like its a known hardware chipset problem, and was
>>>> first reported 6-months ago.
>>>>
>>>> It affects all PCI cards in Asus Sandy-Bridge Motherboards. No known
>>>> fix as of yet.
>>>>
>>>> https://lkml.org/lkml/2012/1/30/216
>>>
>>> At least our discussion got you looking in multiple directions, one of
>>> which led you to this information.
>>>
>>> Given the problem is related to legacy PCI INTx sharing/routing, whether
>>> on the PCI or PCIe bus, I'd recommend you step up to a high quality PCIe
>>> x8 SAS/SATA HBA, such as the LSI 9211-8i PCIe x8, which supports MSI-X
>>> and should instantly solve your problem.
>>> http://www.newegg.com/Product/Product.aspx?Item=N82E16816118112
>>>
>>> You'll need two breakout cables:
>>> http://www.newegg.com/Product/Product.aspx?Item=N82E16816116098
>>>
>>> This solution will set you back almost $300 USD.  I just did some
>>> research on the Syba 4 port SiI 3124 PCIe x1 card.  The SiI 3124 is a
>>> native PCI/X chip, thus the board uses a PCI-X to PCIe bridge chip which
>>> hides under the large heatsink.  Thus this card will not work, as it
>>> uses legacy PCI interrupts:
>>> http://www.newegg.com/Product/Product.aspx?Item=N82E16816124027
>>>
>>> I also looked at the Syba and Adaptec Marvell PCIe x1 SATAII 4 port
>>> cards.  While the Marvell chip is native PCIe I'm unable to confirm it
>>> supports MSI/X.  And given these cards run $80-90, that's $160-180 for
>>> two of them.  The LSI above is pretty much guaranteed to work for ~$100
>>> more.  What's your reputation with your client worth?
>>>
>>> Speaking of which, don't even look at the $110 8 port Supermicro
>>> SAS/SATA controller.  It uses the Marvell SAS chip.  Although the chip
>>> itself is fine and works with Windows, the Linux driver *will* eat your
>>> data, all the way up to kernel 3.4.  I've personally rectified this
>>> situation for a half dozen users who bought this SM SAS board on price
>>> alone.  I converted them all to LSI HBAs and no problems since.  The
>>> solution cost them 2-3x as much but they're all happy because it simply
>>> works reliably, and fast.
>>>
>>> Or you can start swapping $150+ motherboards until you find one that
>>> works with those $20 Syba 3114 cards.  But then you need to ask
>>> yourself, how much is your time worth.  You could easily burn 20 or more
>>> hours going that route.
>>>
>>> Get the LSI and be done with this.
>>>
>>
>> The above sounds like excellent advice, and you have saved me several
>> hours of perusing catalogs trying to figure out what to buy to replace
>> the two SiI cards. I greatly appreciate the help, and I have sent off
>> an order to NewEgg for the LSI board and cables.
>>
> 
> Just wanted to let everyone know that the new hardware arrived
> yesterday, and today my two raid's finished rebuilding without any
> problems at all, and at a fair bit higher speed than before.
> 
> So, thanks for all your help, in getting my backup system operational.

That's great to hear Stirling.  Glad it's working well.  Coincidentally,
an OP posted today to the LInux-RAID list that he was getting silent
data corruption, again, from his second set of cheapo cards, first Syba,
then Rosewill.  I gave him the same advice.

When one spends more on a couple of large pizzas or a case of beer than
on one's disk controller...need I say more?

-- 
Stan