Bogus "igb1: Could not setup receive structures" in 8-STABLE

[Jack Vogel]

The number of MSIX vectors it uses is the number of queues PLUS
one vector for link. I would use two or four rather than 3, but it should
be ok with that if that's what you wish.

Jack


On Thu, Oct 14, 2010 at 6:58 PM, Terry Kennedy <TERRY at tmk.com> wrote:

> > The problem is mbuf resources, the driver is autoconfiguring the number
> of
> > queues based on the number of cores, on newer systems with lots of them
> > this is outstripping the mbuf resource pool.
>
>   That would make sense, as these systems have 16 cores (dual E5520's).
>
> > I have decided to hard limit the queues to 8, you can fix the number
> > manually
> > by searching for num_queues in if_igb.c and setting it to something other
> > than
> > 0 for now.
>
>   I changed it to 8, and saw the same problem. I noted that the igb boot
> messages changed from:
>
> Oct 14 18:28:02 rz1m kernel: igb0: Using MSIX interrupts with 10 vectors
> Oct 14 18:28:02 rz1m kernel: igb1: Using MSIX interrupts with 10 vectors
>
>  to:
>
> Oct 14 21:53:44 rz1m kernel: igb0: Using MSIX interrupts with 9 vectors
> Oct 14 21:53:44 rz1m kernel: igb1: Using MSIX interrupts with 9 vectors
>
>  So I dropped the value to 3 (on the assumption that the system uses one
> more than the specified value per interface), and got:
>
> igb0: Using MSIX interrupts with 4 vectors
> igb1: Using MSIX interrupts with 4 vectors
>
>  and both igb interfaces came up. I didn't try to find the maximum
> number of queues that would work.
>
> > I am at work on a number of issues with igb and em right now which is why
> > there has not been an MFC yet.
>
>   Understood. Thanks for the quick response and workaround.
>
>        Terry Kennedy             http://www.tmk.com
>        terry at tmk.com             New York, NY USA
>