Device timeout from mfi(9) while booting 9.2-RELEASE

Jan Mikkelsen janm at transactionware.com
Mon Oct 7 10:02:08 UTC 2013


On 7 Oct 2013, at 8:06 pm, Steven Hartland <killing at multiplay.co.uk> wrote:

> ----- Original Message ----- From: "Jan Mikkelsen" <janm at transactionware.com>
> ..
> 
>> Just to confirm when you say fine, have you checked to ensure none
>> are reporting timeouts?
> 
>> Yes.
>> Only the first two of these are from a 9.2 era system (also the ones
>> with the most recent firmware).
>> We did have timeout issues while testing this platform which were
>> resolved by taking the mfi driver from -HEAD at the time and setting
>> hw.mfi.msi=1. The oldest ones have been running for over two years.
> 
> Thats curious, hw.mfi.msi=1 is the default so you shouldn't need to
> change this.
> 
> Also heads mfi is almost identical to 9.2 so not sure why you would
> need to update, could you clarify?


The 9.2 system has the 9.2 driver and was not updated. It is a test system with a 9261-8i and a 9240-4i.

The other machines are running 9.0 or 9.1 with a modified driver. The 9.0 systems have a patch that added the hw.mfi.msi sysctl with a default of zero. I just went back to our Perforce depot and checked the timing; that patch was applied to our local copy on 2011-12-08. My “running for over two years” was wrong, it is actually about a year and a half. At that point the 9261-8i was stable and we saw no further timeout errors even with the old firmware versions in my email. (looking through svn, this change was in rev 227562 to mfi_pci.c.)

The loader.conf entry has been in our standard builds ever since, and even after the default value changed. Probably time to remove it.

We started using the 9240-4i/8i in May 2012. We brought in changes from head mfi to support them. I’d need to go digging deeper to see exactly what the changes were. A quick look shows that this is when the hw.mfi.msi default changed in our copy of the code.

In any case: These system are running and we don’t see timeout errors with the firmware versions I listed.

There were also your later changes that improved the reliability of mfi. We brought those back into our 9.1 based systems, but the cards were working before those changes.

Regards,

Jan.



More information about the freebsd-stable mailing list