Data corruption with the mfi(4) driver

[Steven Hartland]

----- Original Message ----- 
From: "Francois Tigeot" <ftigeot at wolfpond.org>

> On Thu, Jul 10, 2014 at 11:20:38AM +0100, Steven Hartland wrote:
>> I cant see any information on the actual corruption or cause in that linked
>> thread do you have any actual details?
>> 
>> There was known corruption issues but these where fixed long ago so would
>> be good to confirm the details of what you where running and the HW when you
>> had the issue.
>> 
>> As a point of reference we have mfi backed DB machines here and have not
>> had any issues with corruption and they have been in production for over
>> 1 1/2 years.
> 
> It is only visible with recent adapters like the Thunderbolt serie, and then
> under relatively high disk load.
> 
> The whole Dell Rx20 generation of servers seem to be impacted; the previous
> Rx10 generation is safe.
> 
> This bug report contains additional details as well as PCI ids from two
> different Dell machines having experienced filesystem destruction:
> http://bugs.dragonflybsd.org/issues/2683
> 
> HAMMER CRC32 errors were reported on the console and the kernel eventually
> crashed after some time; I didn't get crash dumps.

That PCI ids is 2208 based = Thunderbolt for which we have quite a few fixes.

As I mentioned, I'm not sure if dragonflybsd have our mfi fixes for those cards,
so would be good to confirm this along with confirming that the report was
after said fixes.

Boris is using Invader not Thunderbolt so could well be totally different issue.

Also fw can play a key roll in issues like this, so without knowing why and
hence having a fix its impossible to tell if the two reports are related at
all.

    Regards
    Steve