[ofa-general] sminfo report iberror in the first configuration on RHEL5.3

[Doug Ledford]

On Fri, 2009-02-13 at 08:05 +0800, Wen Hao Wang wrote:
> Doug Ledford <dledford at redhat.com> 写于 2009-02-12 21:20:30:
> 
> > On Thu, 2009-02-12 at 13:20 +0200, Tziporet Koren wrote:
> > > Wen Hao Wang wrote:
> > > >
> > > > Hi all:
> > > >
> > > > I changed my blade OS to RHEL5.3 yesterday and installed OFED
> (shipped 
> > > > in RHEL5.3 image) by "yum groupisntall". Then I load some
> drivers and 
> > > > wrote network interface configuration file ifcfg-ib0. ifup ib0
> also 
> > > > succeeded. But IB utilites report Connetion timed out.
> > > >
> > > >
> > > > [root at xblade06 network-scripts]# sminfo
> > > > ibwarn: [32593] _do_madrpc: recv failed: Connection timed out
> > > > ibwarn: [32593] mad_rpc: _do_madrpc failed; dport (Lid 9)
> > > > sminfo: iberror: failed: query
> > > >
> > > > I had to reboot the blade and rerun "openibd start". Then
> sminfo 
> > > > reported correct contents. I do not suppose this reboot is
> required. 
> > > > Did I miss any configuration step?
> > 
> > There was an unintentional bug in the rhel5.2 openibd init script in
> > that it automatically turned itself on during install (generally,
> most
> > init scripts should default to *not* turning themselves on during
> > install of the package, nor should they start themselves during
> install
> > of the package...this is for security reasons, imagine if you
> installed
> > the bind name server on your box and it automatically started up
> before
> > you had a chance to configure it).  In rhel5.3 we fixed that bug.
>  So,
> 
> Yeah. I heard of this bug.
> 
> > you may need to 'chkconfig --level 2345 openibd on' to make sure
> openibd
> > starts up each time.  The error you list above is consistent with
> not
> > all of the kernel modules being loaded when you tried to use the
> sminfo
> > program.
> 
> Even after reboot, service openibd is not started automatically.
> [root at xblade06 ~]# chkconfig --list openibd
> openibd         0:off   1:off   2:off   3:off   4:off   5:off   6:off

That's because you have to run the command I listed in my first email to
turn it on.

> I agree with you that maybe some modules were not loaded. But what's
> that?
> Before reboot, I run "/etc/init.d/openibd start" and
> "/etc/init.d/network
> restart". No error was reported. "openibd status" also looked good.

Running start on a service does not enable that service at the next
reboot.  You must specifically enable the service in order for it to
start automatically.

> > 
> > > > Moreover, "openibd start" report one warning message about
> hwconf. 
> > > > Anyone has comments about this?
> > > >
> > > > [root at xblade07 ~]# /etc/init.d/openibd start
> > > > Loading OpenIB kernel modules:grep: /etc/sysconfig/hwconf: No
> such 
> > > > file or directory
> > > > [ OK ]
> > 
> > Can you see if the kudzu package is installed on your machine?  The
> > openib package uses this config file written by kudzu to determine
> what
> > hardware drivers to load.  I suppose I should put a specific
> requires in
> > the rpm for that.
> 
> kudzu is installed.
> [root at xblade06 ~]# rpm -q kudzu
> kudzu-1.2.57.1.21-1

Make sure kudzu has been run at least once then (it would appear to be
turned off on your machine or else /etc/sysconfig/hwconf would exist).
You can run it manually from the command line and that should be
sufficient for the openibd init script's needs.

-- 
Doug Ledford <dledford at redhat.com>
              GPG KeyID: CFBFF194
              http://people.redhat.com/dledford

Infiniband specific RPMs available at
              http://people.redhat.com/dledford/Infiniband

-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 197 bytes
Desc: This is a digitally signed message part
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20090213/5cc773af/attachment.sig>