Re: kern/40462: bnx0: Double mbuf allocation failure!

["Michael L. Hitch" <mhitch%lightning.msu.montana.edu@localhost>]

The following reply was made to PR kern/40462; it has been noted by GNATS.

From: "Michael L. Hitch" <mhitch%lightning.msu.montana.edu@localhost>
To: gnats-bugs%NetBSD.org@localhost
Cc: kern-bug-people%netbsd.org@localhost, gnats-admin%netbsd.org@localhost, 
netbsd-bugs%netbsd.org@localhost
Subject: Re: kern/40462: bnx0: Double mbuf allocation failure!
Date: Fri, 23 Jan 2009 12:20:36 -0700 (MST)

 On Fri, 23 Jan 2009, jarle%uninett.no@localhost wrote:
 
 >> Description:
 > The system is an IBM e-series rack server with two built-in ethernet
 > ports.  The ports are driven by the bnx driver.  It had been running for
 > some time using only one network port, and I was trying to bring up the
 > second port when it paniced.  I gathered some console log output, and tried
 > again.  Same panic.
 ..
 >> How-To-Repeat:
 > Bring up two bnx interface on a system?
 >
 >> Fix:
 
    I think this occurs because each bnx device attempts to allocate a large
 number of receive buffers from the mbuf cluster pool.  The default size
 is too small for two bnx devices, and the bnx driver appears to be the 
 only one I've seen that panics when it can't allocate the receive buffers.
 
    You can override this with the NMBCLUSTERS kernel config option (which
 required rebuilding the kernel).  Also, enabling the GATEWAY option will
 usually double the default NMBCLUSTERS (but also requires a kernel 
 rebuild).
 
    I think in the past I've been able to patch the nmbclusters value in
 the kernel using gdb.
 
    I also seem to recall some comments or discussion about increasing the 
 default NMBCLUSTERS, but I don't think anything ever came of that.
 
 --
 Michael L. Hitch                       mhitch%montana.edu@localhost
 Computer Consultant
 Information Technology Center
 Montana State University       Bozeman, MT     USA