[openib-general] Re: multcast join failed

[Hal Rosenstock]

On Wed, 2006-05-17 at 15:15, Michael S. Tsirkin wrote:
> Quoting r. Hal Rosenstock <halr at voltaire.com>:
> > Subject: Re: multcast join failed
> > 
> > On Wed, 2006-05-17 at 14:23, Michael S. Tsirkin wrote:
> > > Quoting r. Hal Rosenstock <halr at voltaire.com>:
> > > > Subject: Re: multcast join failed
> > > > 
> > > > On Wed, 2006-05-17 at 11:51, Michael S. Tsirkin wrote:
> > > > > Hi, Roland!
> > > > > With svn trunk, I started getting the following on one machine:
> > > > > 
> > > > > ib0: multicast join failed for ff12:401b:ffff:0:0:0:ffff:ffff, status -22
> > > > > 
> > > > > and I can't ping this machine over ipoib.
> > > > > Any idea?
> > > > 
> > > > What SM are you using ?
> > > > 
> > > > If OpenSM, are there any errors in the osm.log ?
> > > > 
> > > > -- Hal
> > > > 
> > > 
> > > 
> > > opensm
> > > I see these
> > > 
> > > May 17 21:21:05 762211 [41001960] -> osm_mcmr_rcv_join_mgrp: ERR 1B12:
> > > __validate_more_comp_fields, __validate_port_caps, or JoinState = 0 failed,
> > > sending IB_SA_MAD_STATUS_REQ_INVALID
> > > May 17 21:21:21 641961 [41802960] -> osm_mcmr_rcv_join_mgrp: ERR 1B12:
> > > __validate_more_comp_fields, __validate_port_caps, or JoinState = 0 failed,
> > > sending IB_SA_MAD_STATUS_REQ_INVALID
> > > May 17 21:21:21 763080 [42804960] -> osm_mcmr_rcv_join_mgrp: ERR 1B12:
> > > __validate_more_comp_fields, __validate_port_caps, or JoinState = 0 failed,
> > > sending IB_SA_MAD_STATUS_REQ_INVALID
> > > May 17 21:21:37 642611 [41001960] -> osm_mcmr_rcv_join_mgrp: ERR 1B12:
> > > __validate_more_comp_fields, __validate_port_caps, or JoinState = 0 failed,
> > > sending IB_SA_MAD_STATUS_REQ_INVALID
> > > May 17 21:21:37 763955 [41802960] -> osm_mcmr_rcv_join_mgrp: ERR 1B12:
> > > __validate_more_comp_fields, __validate_port_caps, or JoinState = 0 failed,
> > > sending IB_SA_MAD_STATUS_REQ_INVALID
> > 
> > Are you attempting to join a 4x group from a 1x port (or perhaps there
> > is a MTU mismatch between the port and the group) ?
> > 
> > -- Hal
> > 
> 
> Yes, for some reason it came up 1x. but why?

If LinkWidthEnabled on both the HCA and switch port indicate 1x/4x, it
is an autonegotiation thing. Perhaps you have a bad cable ?

-- Hal