[ofa-general] Re: saquery hangs/timeouts

[Hal Rosenstock]

On 11/10/07, Troy Benjegerdes <troy at scl.ameslab.gov> wrote:
> saquery apparently very much dislikes having two ports active at once.
> If I pull the cable off the second port it works.

It's indicating  a timeout querying the SA (for node records which is
the default query). What SM/SA ? Can you provide an ibnetdiscover
output of the topology ?

Does it always work without the -d (with both ports plugged in) ?

Unfortunately, I don't have a machine on which to look at this right
now but perhaps it can be looked at in simulation. How critical is
this ?

-- Hal

> Troy Benjegerdes wrote:
> > What reasons could cause the following:
> >
> > [root at sm1 infiniband]# saquery -d
> > Nov 09 22:30:00 692541 [C94EDFB0] -> osm_vendor_bind: Binding to port
> > 0x2c9021b701236
> > Nov 09 22:30:04 779705 [41001940] -> umad_receiver: ERR 5409: send
> > completed with error (method=0x12 attr=0x11 trans_id=0x6400000001) --
> > dropping
> > Nov 09 22:30:04 779716 [41001940] -> umad_receiver: ERR 5410: class
> > 0x3 LID 0x12
> > Query SA failed: IB_TIMEOUT
> >
> > This occurs on a machine which has had both a mthca and mlx4 card, and
> > an almost identical machine with another mlx4 card works just fine.
> >
> > The only real difference I can tell is that the machine that works
> > previously had OFED-1.3 alpha 1 installed, and the one that does not
> > work has not had OFED-1.3 installed. I also get the hang on my debian
> > systems that I built the kernel, libibverbs, libmthca, etc myself.
> >
> > The debian system gets the following behavior:
> >
> > bash-3.1# /opt/sc07/sbin/saquery -d
> > Nov 09 22:36:14 336986 [F7EDD6C0] -> osm_vendor_bind: Binding to port
> > 0x2c90300001dd1
> > NodeRecord dump:
> >                lid.....................0xA1
> >                reserved................0x0
> >                base_version............0x
> >
>
> _______________________________________________
> general mailing list
> general at lists.openfabrics.org
> http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general
>
> To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
>