[openib-general] Re: OpenSM 1.8.0 Merge Status and Operational Issue

[Hal Rosenstock]

On Fri, 2005-09-02 at 02:11, Eitan Zahavi wrote:
> This was quick. I guess Yael's work on merging the 1.8.0 against 
> the truck did help.

Yes, Yael's hard work made this easy for me.

> > I have a 4x HCA port (1x/4x LinkWidthEnable and Supported) connected via
> > a 1x analyzer connected to a switch (so is 1x LinkWidthActive).
> > OpenSM does not seem to want to bring this port up. It tries once and
> > gives up until the physical link is cycled (cable pull and reinsertion).
> > It does work running over a 4x link with 4x neighbor ports.
> Can you try this with the pre-merge build? Does it work? Just to make sure it
> is a new bug.

It works with the old OpenIB OpenSM (at least it will bring the port
back if it doesn't come up the first time).

> Also with the 1.8.0 you could force the SM to heavy sweep by sending it
> kill -HUP

Yes, that makes it go through the cycle again but it still doesn't come
up.

> Please provide detailed (-V) log file so we can see what is happening too.

Attached is a gzip'd version of the log. The port in question is GUID
0x0008f10403960559.

There are a number of errors detected on Sets of PortInfo which is 
consistent with what I saw on the wire.

I also see these all over the log:
Sep 02 15:27:02 548378 [B572BA40] -> osm_state_mgr_process: Received signal OSM_SIGNAL_IDLE_TIME_PROCESS_REQUEST in state OSM_SM_STATE_SET_ACTIVE_WAIT.
Sep 02 15:27:02 548398 [B572BA40] -> __osm_state_mgr_signal_warning: Invalid signal OSM_SIGNAL_IDLE_TIME_PROCESS_REQUEST(9) in state OSM_SM_STATE_SET_ACTIVE_WAIT.

> > Can you try this ?
> We will on Sunday/Monday. I have a feeling I run into this once. Not sure if it is not an Analyzer issue.
> Could you please take the Analyzer out and retry?

Unfortunately I can't and keep it at 1x as I don't have a cable to do
this.

-- Hal
-------------- next part --------------
A non-text attachment was scrubbed...
Name: osm.log.gz
Type: application/x-gzip
Size: 181007 bytes
Desc: not available
URL: <http://lists.openfabrics.org/pipermail/general/attachments/20050902/08dd9815/attachment.bin>